https://jcollopy-tulane.github.io

Exploring Decision Fights in the UFC

- John Collopy

Links to Each Section

1. Motivation
2. Further Reading
3. ETL
4. EDA
5. Model Selection
6. Conclusions

Motivation

I find Mixed Martial Arts compelling because of the unpredictability in competition. Danger is always imminent, and an exciting ending is always on the table. That being said, a large number of fights do not end in a decisive knockout or submission, and so three judges are left to decide who won the fight. The three judges are supposed to judge the fight according to the Unified Rules of Mixed Arts.

Judges assess the fight according to three criteria:

• Effective striking/grappling
• Effective aggressiveness
• Control of the fighting area

However, this language is subjective and leaves room for interpretation. Thus, perhaps unsurprisingly, fans and fighters find themselves frustrated with how judges decide fights.

The goal of this project is to use fight data to predict the results of fights and assess the impact of fighter output. In doing so, we will gain a better understand of the criteria by which judges score fights.

Further Reading

For those interested in learning more about the judging criteria, controversial decisions, or frustrated athletes, a few links are provided below.

• Quick and Easy Guide to MMA Judging Criteria
• Controversial Decisions in UFC History
• Athlete Upset With Judges After Controversial Loss

ETL

The data used for this project comes from ufcstats.com, a website containing data from every UFC fight. This site contains multiple "layers", from which separate datasets were created. The first "layer" of this site contains links that track simple counting statistics for each fight, such as number of knockdowns, significant strikes, takedowns, and submission attempts. This data will assist in the initial EDA, as I will create charts displaying the distibutions of the winner's differentials with respect to these features. The second "layer" of this site contains more advanced metrics about each fight. This data will be merged with the first dataset in order to build a more detailed model that will predict fight results.

Important Note: This dataset contains all UFC fights up to and including the Islam Makhachev vs. Alexander Volkanovski match, which took place on October 21, 2023.

# Importing the necessary packages to scrape the data

import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from bs4 import BeautifulSoup as bs
import requests
from tqdm import tqdm
import re

# Retrieving the URL that contains links to all fight data

url = "http://www.ufcstats.com/statistics/events/completed?page=all"
headers = {"User-Agent": "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/605.1.15 (KHTML, like Gecko) Version/16.3 Safari/605.1.15"}


html = requests.get(url, headers = headers)
soup = bs(html.content, "html.parser")

I now have the content for the site, and I will put the link containing data for each fight into a list, which will then be scraped for fight data.

Scraping the First "Layer" of Data

# Putting each link on the home page into a list

first_layer = []
for link in tqdm(soup.find_all('a', href=True), desc="Scraping First Layer"):
    page = link['href']
    first_layer.append(page)

Scraping First Layer: 100%|███████████████| 694/694 [00:00<00:00, 293698.62it/s]

# Removing all links that do not contain relevant fight data

first_layer = first_layer[11:-16] 

I now have some preliminary data for each fight. The data must be cleaned and requires some feature engineering, but since the data took some time to scrape, I will now save the dataframe as a .csv file, which will be imported later.

# Extracting Data From the "first layer" of the UFC Stats site and creating a dataframe

main_dataframe = pd.DataFrame()

for url in tqdm(first_layer, desc="Scraping First Layer"):
    html = requests.get(url, headers=headers)
    soup = bs(html.content, "html.parser")
    table = soup.find("table")
    data = pd.read_html(str(table))
    adj_data = pd.concat(data)
    
    event = soup.find("h2", class_="b-content__title").text.strip()
    

    adj_data["Event"] = event
    
    main_dataframe = pd.concat([main_dataframe, adj_data], ignore_index=True)
    
main_dataframe.head()

Scraping First Layer: 100%|███████████████████| 667/667 [04:59<00:00,  2.22it/s]

	W/L	Fighter	Kd	Str	Td	Sub	Weight class	Method	Round	Time	Event
0	win	Islam Makhachev Alexander Volkanovski	1 0	24 4	0 0	0 0	Lightweight	KO/TKO Kick	1	3:06	UFC 294: Makhachev vs. Volkanovski 2
1	win	Khamzat Chimaev Kamaru Usman	0 0	38 36	4 0	1 0	Middleweight	M-DEC	3	5:00	UFC 294: Makhachev vs. Volkanovski 2
2	nc nc	Magomed Ankalaev Johnny Walker	0 0	13 14	1 0	0 0	Light Heavyweight	CNC	1	3:13	UFC 294: Makhachev vs. Volkanovski 2
3	win	Ikram Aliskerov Warlley Alves	1 0	26 12	0 0	0 0	Middleweight	KO/TKO Punches	1	2:07	UFC 294: Makhachev vs. Volkanovski 2
4	win	Said Nurmagomedov Muin Gafurov	0 0	5 1	0 0	1 0	Bantamweight	SUB Guillotine Choke	1	1:13	UFC 294: Makhachev vs. Volkanovski 2

main_dataframe.to_csv('ufc_initial_file.csv')

Scraping the Second Layer of Data (Part 1)

# Getting the links for each fight

second_layer = []

with tqdm(first_layer, desc="Scraping Links") as layer:
    for link in layer:
        html = requests.get(link, headers = headers)
        soup = bs(html.content, "lxml")
        
        for link in soup.find_all('a', href=True):
            if "fight-details" in link['href']:
                    second_layer.append(link['href'])

Scraping Links: 100%|█████████████████████████| 667/667 [04:40<00:00,  2.37it/s]

# Creating a dataframe and scraping the data

main_df_1 = pd.DataFrame()  

for url in tqdm(second_layer[:1000], desc="Scraping Second Layer"):
    html = requests.get(url, headers=headers)
    soup = bs(html.content, "lxml")
    tables = soup.find_all("table")
    table_dataframes = []

    
    for table in tables:
        table_df = pd.read_html(str(table))[0] 
        table_dataframes.append(table_df)  

   
    event_name = soup.find("h2", class_="b-content__title").text.strip()

    combined_table = pd.concat(table_dataframes, ignore_index=True)[["Fighter", "Sig. str.", "Td", "Rev.", "Ctrl"]].dropna()

    
    combined_table["Event"] = event_name

    main_df_1 = pd.concat([main_df_1, combined_table], ignore_index=True)

main_df_1.head()

Scraping Second Layer: 100%|████████████████| 1000/1000 [11:49<00:00,  1.41it/s]

	Fighter	Sig. str.	Td	Rev.	Ctrl	Event
0	Islam Makhachev Alexander Volkanovski	24 of 29 4 of 6	0 of 1 0 of 0	0 0	0:19 0:49	UFC 294: Makhachev vs. Volkanovski 2
1	Kamaru Usman Khamzat Chimaev	36 of 66 38 of 70	0 of 1 4 of 12	0 0	0:04 7:16	UFC 294: Makhachev vs. Volkanovski 2
2	Magomed Ankalaev Johnny Walker	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00	UFC 294: Makhachev vs. Volkanovski 2
3	Magomed Ankalaev Johnny Walker	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00	UFC 294: Makhachev vs. Volkanovski 2
4	Ikram Aliskerov Warlley Alves	26 of 36 12 of 16	0 of 0 0 of 0	0 0	0:00 0:00	UFC 294: Makhachev vs. Volkanovski 2

# Continued scraping

for url in tqdm(second_layer[1000:2000], desc="Scraping Second Layer"):
    html = requests.get(url, headers=headers)
    soup = bs(html.content, "lxml")
    tables = soup.find_all("table")
    table_dataframes = []

    
    for table in tables:
        table_df = pd.read_html(str(table))[0] 
        table_dataframes.append(table_df)  

   
    event_name = soup.find("h2", class_="b-content__title").text.strip()

    combined_table = pd.concat(table_dataframes, ignore_index=True)[["Fighter", "Sig. str.", "Td", "Rev.", "Ctrl"]].dropna()

    
    combined_table["Event"] = event_name

    main_df_1 = pd.concat([main_df_1, combined_table], ignore_index=True)

main_df_1.head()

Scraping Second Layer: 100%|████████████████| 1000/1000 [11:46<00:00,  1.42it/s]

	Fighter	Sig. str.	Td	Rev.	Ctrl	Event
0	Islam Makhachev Alexander Volkanovski	24 of 29 4 of 6	0 of 1 0 of 0	0 0	0:19 0:49	UFC 294: Makhachev vs. Volkanovski 2
1	Kamaru Usman Khamzat Chimaev	36 of 66 38 of 70	0 of 1 4 of 12	0 0	0:04 7:16	UFC 294: Makhachev vs. Volkanovski 2
2	Magomed Ankalaev Johnny Walker	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00	UFC 294: Makhachev vs. Volkanovski 2
3	Magomed Ankalaev Johnny Walker	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00	UFC 294: Makhachev vs. Volkanovski 2
4	Ikram Aliskerov Warlley Alves	26 of 36 12 of 16	0 of 0 0 of 0	0 0	0:00 0:00	UFC 294: Makhachev vs. Volkanovski 2

# Continued scraping

for url in tqdm(second_layer[2000:3000], desc="Scraping Second Layer"):
    html = requests.get(url, headers=headers)
    soup = bs(html.content, "lxml")
    tables = soup.find_all("table")
    table_dataframes = []

    
    for table in tables:
        table_df = pd.read_html(str(table))[0] 
        table_dataframes.append(table_df)  

   
    event_name = soup.find("h2", class_="b-content__title").text.strip()

    combined_table = pd.concat(table_dataframes, ignore_index=True)[["Fighter", "Sig. str.", "Td", "Rev.", "Ctrl"]].dropna()

    
    combined_table["Event"] = event_name

    main_df_1 = pd.concat([main_df_1, combined_table], ignore_index=True)

main_df_1.head()

Scraping Second Layer: 100%|████████████████| 1000/1000 [12:05<00:00,  1.38it/s]

	Fighter	Sig. str.	Td	Rev.	Ctrl	Event
0	Islam Makhachev Alexander Volkanovski	24 of 29 4 of 6	0 of 1 0 of 0	0 0	0:19 0:49	UFC 294: Makhachev vs. Volkanovski 2
1	Kamaru Usman Khamzat Chimaev	36 of 66 38 of 70	0 of 1 4 of 12	0 0	0:04 7:16	UFC 294: Makhachev vs. Volkanovski 2
2	Magomed Ankalaev Johnny Walker	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00	UFC 294: Makhachev vs. Volkanovski 2
3	Magomed Ankalaev Johnny Walker	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00	UFC 294: Makhachev vs. Volkanovski 2
4	Ikram Aliskerov Warlley Alves	26 of 36 12 of 16	0 of 0 0 of 0	0 0	0:00 0:00	UFC 294: Makhachev vs. Volkanovski 2

# Continued scraping

for url in tqdm(second_layer[3000:4000], desc="Scraping Second Layer"):
    html = requests.get(url, headers=headers)
    soup = bs(html.content, "lxml")
    tables = soup.find_all("table")
    table_dataframes = []

    
    for table in tables:
        table_df = pd.read_html(str(table))[0] 
        table_dataframes.append(table_df)  

   
    event_name = soup.find("h2", class_="b-content__title").text.strip()

    combined_table = pd.concat(table_dataframes, ignore_index=True)[["Fighter", "Sig. str.", "Td", "Rev.", "Ctrl"]].dropna()

    
    combined_table["Event"] = event_name

    main_df_1 = pd.concat([main_df_1, combined_table], ignore_index=True)

main_df_1.head()

Scraping Second Layer: 100%|████████████████| 1000/1000 [12:07<00:00,  1.37it/s]

	Fighter	Sig. str.	Td	Rev.	Ctrl	Event
0	Islam Makhachev Alexander Volkanovski	24 of 29 4 of 6	0 of 1 0 of 0	0 0	0:19 0:49	UFC 294: Makhachev vs. Volkanovski 2
1	Kamaru Usman Khamzat Chimaev	36 of 66 38 of 70	0 of 1 4 of 12	0 0	0:04 7:16	UFC 294: Makhachev vs. Volkanovski 2
2	Magomed Ankalaev Johnny Walker	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00	UFC 294: Makhachev vs. Volkanovski 2
3	Magomed Ankalaev Johnny Walker	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00	UFC 294: Makhachev vs. Volkanovski 2
4	Ikram Aliskerov Warlley Alves	26 of 36 12 of 16	0 of 0 0 of 0	0 0	0:00 0:00	UFC 294: Makhachev vs. Volkanovski 2

# Continued scraping

for url in tqdm(second_layer[4000:5000], desc="Scraping Second Layer"):
    html = requests.get(url, headers=headers)
    soup = bs(html.content, "lxml")
    tables = soup.find_all("table")
    table_dataframes = []

    
    for table in tables:
        table_df = pd.read_html(str(table))[0] 
        table_dataframes.append(table_df)  

   
    event_name = soup.find("h2", class_="b-content__title").text.strip()

    combined_table = pd.concat(table_dataframes, ignore_index=True)[["Fighter", "Sig. str.", "Td", "Rev.", "Ctrl"]].dropna()

    
    combined_table["Event"] = event_name

    main_df_1 = pd.concat([main_df_1, combined_table], ignore_index=True)

main_df_1.head()

Scraping Second Layer: 100%|████████████████| 1000/1000 [12:50<00:00,  1.30it/s]

	Fighter	Sig. str.	Td	Rev.	Ctrl	Event
0	Islam Makhachev Alexander Volkanovski	24 of 29 4 of 6	0 of 1 0 of 0	0 0	0:19 0:49	UFC 294: Makhachev vs. Volkanovski 2
1	Kamaru Usman Khamzat Chimaev	36 of 66 38 of 70	0 of 1 4 of 12	0 0	0:04 7:16	UFC 294: Makhachev vs. Volkanovski 2
2	Magomed Ankalaev Johnny Walker	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00	UFC 294: Makhachev vs. Volkanovski 2
3	Magomed Ankalaev Johnny Walker	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00	UFC 294: Makhachev vs. Volkanovski 2
4	Ikram Aliskerov Warlley Alves	26 of 36 12 of 16	0 of 0 0 of 0	0 0	0:00 0:00	UFC 294: Makhachev vs. Volkanovski 2

# Continued scraping

for url in tqdm(second_layer[5000:6000], desc="Scraping Second Layer"):
    html = requests.get(url, headers=headers)
    soup = bs(html.content, "lxml")
    tables = soup.find_all("table")
    table_dataframes = []

    
    for table in tables:
        table_df = pd.read_html(str(table))[0] 
        table_dataframes.append(table_df)  

   
    event_name = soup.find("h2", class_="b-content__title").text.strip()

    combined_table = pd.concat(table_dataframes, ignore_index=True)[["Fighter", "Sig. str.", "Td", "Rev.", "Ctrl"]].dropna()

    
    combined_table["Event"] = event_name

    main_df_1 = pd.concat([main_df_1, combined_table], ignore_index=True)

main_df_1.head()

Scraping Second Layer: 100%|████████████████| 1000/1000 [11:53<00:00,  1.40it/s]

	Fighter	Sig. str.	Td	Rev.	Ctrl	Event
0	Islam Makhachev Alexander Volkanovski	24 of 29 4 of 6	0 of 1 0 of 0	0 0	0:19 0:49	UFC 294: Makhachev vs. Volkanovski 2
1	Kamaru Usman Khamzat Chimaev	36 of 66 38 of 70	0 of 1 4 of 12	0 0	0:04 7:16	UFC 294: Makhachev vs. Volkanovski 2
2	Magomed Ankalaev Johnny Walker	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00	UFC 294: Makhachev vs. Volkanovski 2
3	Magomed Ankalaev Johnny Walker	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00	UFC 294: Makhachev vs. Volkanovski 2
4	Ikram Aliskerov Warlley Alves	26 of 36 12 of 16	0 of 0 0 of 0	0 0	0:00 0:00	UFC 294: Makhachev vs. Volkanovski 2

# Continued scraping

for url in tqdm(second_layer[6000:7340], desc="Scraping Second Layer"):
    html = requests.get(url, headers=headers)
    soup = bs(html.content, "lxml")
    tables = soup.find_all("table")
    table_dataframes = []

    
    for table in tables:
        table_df = pd.read_html(str(table))[0] 
        table_dataframes.append(table_df)  

   
    event_name = soup.find("h2", class_="b-content__title").text.strip()

    combined_table = pd.concat(table_dataframes, ignore_index=True)[["Fighter", "Sig. str.", "Td", "Rev.", "Ctrl"]].dropna()

    
    combined_table["Event"] = event_name

    main_df_1 = pd.concat([main_df_1, combined_table], ignore_index=True)

main_df_1.head()

Scraping Second Layer: 100%|████████████████| 1340/1340 [13:18<00:00,  1.68it/s]

	Fighter	Sig. str.	Td	Rev.	Ctrl	Event
0	Islam Makhachev Alexander Volkanovski	24 of 29 4 of 6	0 of 1 0 of 0	0 0	0:19 0:49	UFC 294: Makhachev vs. Volkanovski 2
1	Kamaru Usman Khamzat Chimaev	36 of 66 38 of 70	0 of 1 4 of 12	0 0	0:04 7:16	UFC 294: Makhachev vs. Volkanovski 2
2	Magomed Ankalaev Johnny Walker	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00	UFC 294: Makhachev vs. Volkanovski 2
3	Magomed Ankalaev Johnny Walker	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00	UFC 294: Makhachev vs. Volkanovski 2
4	Ikram Aliskerov Warlley Alves	26 of 36 12 of 16	0 of 0 0 of 0	0 0	0:00 0:00	UFC 294: Makhachev vs. Volkanovski 2

main_df_1.to_csv("ufc_second_csv")

Scraping the Second Layer of Data (Part 2)

The second layer of data took a while to scrape, and so I will save this dataset as a .csv file as well, which will be imported later. Also, due to a number of issues scraping all of the data at once, I broke the scraping up into bits to ensure that the code ran to completion.

# Initialize the main DataFrame

main_dataframe_1 = pd.DataFrame()  

for url in tqdm(second_layer[:500], desc="Scraping Second Layer"):
    html = requests.get(url, headers=headers)
    soup = bs(html.content, "lxml")
    tables = soup.find_all("table")
    table_dataframes = []

    
    for table in tables:
        table_df = pd.read_html(str(table))[0] 
        table_dataframes.append(table_df)  

   
    event_name = soup.find("h2", class_="b-content__title").text.strip()

    combined_table = pd.concat(table_dataframes, ignore_index=True)[["Fighter", "Head", "Body", "Leg", "Distance", "Clinch", "Ground"]].dropna()

    
    combined_table["Event"] = event_name

    main_dataframe_1 = pd.concat([main_dataframe_1, combined_table], ignore_index=True)

main_dataframe_1.head()

Scraping Second Layer: 100%|██████████████████| 500/500 [06:17<00:00,  1.33it/s]

	Fighter	Head	Body	Leg	Distance	Clinch	Ground	Event
0	Islam Makhachev Alexander Volkanovski	13 of 16 0 of 1	9 of 11 1 of 1	2 of 2 3 of 4	7 of 11 3 of 5	8 of 9 1 of 1	9 of 9 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
1	Kamaru Usman Khamzat Chimaev	27 of 52 27 of 53	6 of 10 3 of 9	3 of 4 8 of 8	33 of 63 22 of 53	1 of 1 0 of 0	2 of 2 16 of 17	UFC 294: Makhachev vs. Volkanovski 2
2	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
3	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
4	Ikram Aliskerov Warlley Alves	20 of 29 3 of 6	6 of 6 1 of 1	0 of 1 8 of 9	26 of 36 12 of 16	0 of 0 0 of 0	0 of 0 0 of 0	UFC 294: Makhachev vs. Volkanovski 2

# Scraping the next 1,500 links

for url in tqdm(second_layer[500:2000], desc="Scraping Second Layer"):
    html = requests.get(url, headers=headers)
    soup = bs(html.content, "lxml")
    tables = soup.find_all("table")
    table_dataframes = []

    
    for table in tables:
        table_df = pd.read_html(str(table))[0] 
        table_dataframes.append(table_df)  

   
    event_name = soup.find("h2", class_="b-content__title").text.strip()

    combined_table = pd.concat(table_dataframes, ignore_index=True)[["Fighter", "Ctrl", "Head", "Body", "Leg", "Distance", "Clinch", "Ground"]].dropna()

    
    combined_table["Event"] = event_name

    main_dataframe_1 = pd.concat([main_dataframe_1, combined_table], ignore_index=True)

main_dataframe_1.head()

Scraping Second Layer: 100%|████████████████| 1500/1500 [19:14<00:00,  1.30it/s]

	Fighter	Head	Body	Leg	Distance	Clinch	Ground	Event
0	Islam Makhachev Alexander Volkanovski	13 of 16 0 of 1	9 of 11 1 of 1	2 of 2 3 of 4	7 of 11 3 of 5	8 of 9 1 of 1	9 of 9 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
1	Kamaru Usman Khamzat Chimaev	27 of 52 27 of 53	6 of 10 3 of 9	3 of 4 8 of 8	33 of 63 22 of 53	1 of 1 0 of 0	2 of 2 16 of 17	UFC 294: Makhachev vs. Volkanovski 2
2	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
3	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
4	Ikram Aliskerov Warlley Alves	20 of 29 3 of 6	6 of 6 1 of 1	0 of 1 8 of 9	26 of 36 12 of 16	0 of 0 0 of 0	0 of 0 0 of 0	UFC 294: Makhachev vs. Volkanovski 2

# Scraping the next 1,000 links

for url in tqdm(second_layer[2000:3000], desc="Scraping Second Layer"):
    html = requests.get(url, headers=headers)
    soup = bs(html.content, "lxml")
    tables = soup.find_all("table")
    table_dataframes = []

    
    for table in tables:
        table_df = pd.read_html(str(table))[0] 
        table_dataframes.append(table_df)  

   
    event_name = soup.find("h2", class_="b-content__title").text.strip()

    combined_table = pd.concat(table_dataframes, ignore_index=True)[["Fighter", "Head", "Body", "Leg", "Distance", "Clinch", "Ground"]].dropna()

    
    combined_table["Event"] = event_name

    main_dataframe_1 = pd.concat([main_dataframe_1, combined_table], ignore_index=True)

main_dataframe_1.head()

Scraping Second Layer: 100%|████████████████| 1000/1000 [13:36<00:00,  1.22it/s]

	Fighter	Head	Body	Leg	Distance	Clinch	Ground	Event
0	Islam Makhachev Alexander Volkanovski	13 of 16 0 of 1	9 of 11 1 of 1	2 of 2 3 of 4	7 of 11 3 of 5	8 of 9 1 of 1	9 of 9 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
1	Kamaru Usman Khamzat Chimaev	27 of 52 27 of 53	6 of 10 3 of 9	3 of 4 8 of 8	33 of 63 22 of 53	1 of 1 0 of 0	2 of 2 16 of 17	UFC 294: Makhachev vs. Volkanovski 2
2	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
3	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
4	Ikram Aliskerov Warlley Alves	20 of 29 3 of 6	6 of 6 1 of 1	0 of 1 8 of 9	26 of 36 12 of 16	0 of 0 0 of 0	0 of 0 0 of 0	UFC 294: Makhachev vs. Volkanovski 2

# Scraping the next 1,000 links

for url in tqdm(second_layer[3000:4000], desc="Scraping Second Layer"):
    html = requests.get(url, headers=headers)
    soup = bs(html.content, "lxml")
    tables = soup.find_all("table")
    table_dataframes = []

    
    for table in tables:
        table_df = pd.read_html(str(table))[0] 
        table_dataframes.append(table_df)  

   
    event_name = soup.find("h2", class_="b-content__title").text.strip()

    combined_table = pd.concat(table_dataframes, ignore_index=True)[["Fighter", "Head", "Body", "Leg", "Distance", "Clinch", "Ground"]].dropna()

    
    combined_table["Event"] = event_name

    main_dataframe_1 = pd.concat([main_dataframe_1, combined_table], ignore_index=True)

main_dataframe_1.head()

Scraping Second Layer: 100%|████████████████| 1000/1000 [13:43<00:00,  1.21it/s]

	Fighter	Head	Body	Leg	Distance	Clinch	Ground	Event
0	Islam Makhachev Alexander Volkanovski	13 of 16 0 of 1	9 of 11 1 of 1	2 of 2 3 of 4	7 of 11 3 of 5	8 of 9 1 of 1	9 of 9 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
1	Kamaru Usman Khamzat Chimaev	27 of 52 27 of 53	6 of 10 3 of 9	3 of 4 8 of 8	33 of 63 22 of 53	1 of 1 0 of 0	2 of 2 16 of 17	UFC 294: Makhachev vs. Volkanovski 2
2	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
3	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
4	Ikram Aliskerov Warlley Alves	20 of 29 3 of 6	6 of 6 1 of 1	0 of 1 8 of 9	26 of 36 12 of 16	0 of 0 0 of 0	0 of 0 0 of 0	UFC 294: Makhachev vs. Volkanovski 2

# Scraping the next 1,000 links

for url in tqdm(second_layer[4000:5000], desc="Scraping Second Layer"):
    html = requests.get(url, headers=headers)
    soup = bs(html.content, "lxml")
    tables = soup.find_all("table")
    table_dataframes = []

    
    for table in tables:
        table_df = pd.read_html(str(table))[0] 
        table_dataframes.append(table_df)  

   
    event_name = soup.find("h2", class_="b-content__title").text.strip()

    combined_table = pd.concat(table_dataframes, ignore_index=True)[["Fighter", "Head", "Body", "Leg", "Distance", "Clinch", "Ground"]].dropna()

    
    combined_table["Event"] = event_name

    main_dataframe_1 = pd.concat([main_dataframe_1, combined_table], ignore_index=True)

main_dataframe_1.head()

Scraping Second Layer: 100%|████████████████| 1000/1000 [13:27<00:00,  1.24it/s]

	Fighter	Head	Body	Leg	Distance	Clinch	Ground	Event
0	Islam Makhachev Alexander Volkanovski	13 of 16 0 of 1	9 of 11 1 of 1	2 of 2 3 of 4	7 of 11 3 of 5	8 of 9 1 of 1	9 of 9 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
1	Kamaru Usman Khamzat Chimaev	27 of 52 27 of 53	6 of 10 3 of 9	3 of 4 8 of 8	33 of 63 22 of 53	1 of 1 0 of 0	2 of 2 16 of 17	UFC 294: Makhachev vs. Volkanovski 2
2	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
3	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
4	Ikram Aliskerov Warlley Alves	20 of 29 3 of 6	6 of 6 1 of 1	0 of 1 8 of 9	26 of 36 12 of 16	0 of 0 0 of 0	0 of 0 0 of 0	UFC 294: Makhachev vs. Volkanovski 2

# Scraping the next 1,000 links

for url in tqdm(second_layer[5000:6000], desc="Scraping Second Layer"):
    html = requests.get(url, headers=headers)
    soup = bs(html.content, "lxml")
    tables = soup.find_all("table")
    table_dataframes = []

    
    for table in tables:
        table_df = pd.read_html(str(table))[0] 
        table_dataframes.append(table_df)  

   
    event_name = soup.find("h2", class_="b-content__title").text.strip()

    combined_table = pd.concat(table_dataframes, ignore_index=True)[["Fighter", "Head", "Body", "Leg", "Distance", "Clinch", "Ground"]].dropna()

    
    combined_table["Event"] = event_name

    main_dataframe_1 = pd.concat([main_dataframe_1, combined_table], ignore_index=True)

main_dataframe_1.head()

Scraping Second Layer: 100%|████████████████| 1000/1000 [13:35<00:00,  1.23it/s]

	Fighter	Head	Body	Leg	Distance	Clinch	Ground	Event
0	Islam Makhachev Alexander Volkanovski	13 of 16 0 of 1	9 of 11 1 of 1	2 of 2 3 of 4	7 of 11 3 of 5	8 of 9 1 of 1	9 of 9 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
1	Kamaru Usman Khamzat Chimaev	27 of 52 27 of 53	6 of 10 3 of 9	3 of 4 8 of 8	33 of 63 22 of 53	1 of 1 0 of 0	2 of 2 16 of 17	UFC 294: Makhachev vs. Volkanovski 2
2	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
3	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
4	Ikram Aliskerov Warlley Alves	20 of 29 3 of 6	6 of 6 1 of 1	0 of 1 8 of 9	26 of 36 12 of 16	0 of 0 0 of 0	0 of 0 0 of 0	UFC 294: Makhachev vs. Volkanovski 2

# Scraping the rest of the links

for url in tqdm(second_layer[6000:7340], desc="Scraping Second Layer"):
    html = requests.get(url, headers=headers)
    soup = bs(html.content, "lxml")
    tables = soup.find_all("table")
    table_dataframes = []

    
    for table in tables:
        table_df = pd.read_html(str(table))[0] 
        table_dataframes.append(table_df)  

   
    event_name = soup.find("h2", class_="b-content__title").text.strip()

    combined_table = pd.concat(table_dataframes, ignore_index=True)[["Fighter", "Head", "Body", "Leg", "Distance", "Clinch", "Ground"]].dropna()

    
    combined_table["Event"] = event_name

    main_dataframe_1 = pd.concat([main_dataframe_1, combined_table], ignore_index=True)

main_dataframe_1.head()

Scraping Second Layer: 100%|████████████████| 1340/1340 [15:49<00:00,  1.41it/s]

	Fighter	Head	Body	Leg	Distance	Clinch	Ground	Event
0	Islam Makhachev Alexander Volkanovski	13 of 16 0 of 1	9 of 11 1 of 1	2 of 2 3 of 4	7 of 11 3 of 5	8 of 9 1 of 1	9 of 9 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
1	Kamaru Usman Khamzat Chimaev	27 of 52 27 of 53	6 of 10 3 of 9	3 of 4 8 of 8	33 of 63 22 of 53	1 of 1 0 of 0	2 of 2 16 of 17	UFC 294: Makhachev vs. Volkanovski 2
2	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
3	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
4	Ikram Aliskerov Warlley Alves	20 of 29 3 of 6	6 of 6 1 of 1	0 of 1 8 of 9	26 of 36 12 of 16	0 of 0 0 of 0	0 of 0 0 of 0	UFC 294: Makhachev vs. Volkanovski 2

main_dataframe_1.to_csv('ufc_compiled.csv')

Unifying the Datasets

Now, it is time to unify and clean the data. Here are the variables in the initial dataframe:

• Unnamed: 0: Index Column
• W/L: Indicates the outcome of the fight
• Fighter: Contains the names of the fighter (Winning fighter on the left)
• Kd: Number of knockdowns for each fighter (Winning fighter's number on the left)
• Str: Number of significant strikes for each fighter (Winning fighter's number on the left)
• Td: Number of takedowns for each fighter (Winning fighter's number on the left)
• Sub: Number of submission attempts for each fighter (Winning fighter's number on the left)
• Weight class: Weight class in which the fight took place
• Method: Method of victory
• Round: Round in which the fight ended
• Time: Time at which the fight ended

I will remove unnecessary columns and remove leading and trailing whitespace in the "Fighter" column. I will also check for null values.

data_1 = pd.read_csv("ufc_initial_file.csv")
data_1.head()

	Unnamed: 0	W/L	Fighter	Kd	Str	Td	Sub	Weight class	Method	Round	Time	Event
0	0	win	Islam Makhachev Alexander Volkanovski	1 0	24 4	0 0	0 0	Lightweight	KO/TKO Kick	1	3:06	UFC 294: Makhachev vs. Volkanovski 2
1	1	win	Khamzat Chimaev Kamaru Usman	0 0	38 36	4 0	1 0	Middleweight	M-DEC	3	5:00	UFC 294: Makhachev vs. Volkanovski 2
2	2	nc nc	Magomed Ankalaev Johnny Walker	0 0	13 14	1 0	0 0	Light Heavyweight	CNC	1	3:13	UFC 294: Makhachev vs. Volkanovski 2
3	3	win	Ikram Aliskerov Warlley Alves	1 0	26 12	0 0	0 0	Middleweight	KO/TKO Punches	1	2:07	UFC 294: Makhachev vs. Volkanovski 2
4	4	win	Said Nurmagomedov Muin Gafurov	0 0	5 1	0 0	1 0	Bantamweight	SUB Guillotine Choke	1	1:13	UFC 294: Makhachev vs. Volkanovski 2

# Dropping Unnecessary columns and changing column names
data_1 = data_1.drop(columns=["Unnamed: 0", "W/L", "Time"], axis=1)
data_1["Fighter"] = data_1["Fighter"].str.strip()
data_1.head()

	Fighter	Kd	Str	Td	Sub	Weight class	Method	Round	Event
0	Islam Makhachev Alexander Volkanovski	1 0	24 4	0 0	0 0	Lightweight	KO/TKO Kick	1	UFC 294: Makhachev vs. Volkanovski 2
1	Khamzat Chimaev Kamaru Usman	0 0	38 36	4 0	1 0	Middleweight	M-DEC	3	UFC 294: Makhachev vs. Volkanovski 2
2	Magomed Ankalaev Johnny Walker	0 0	13 14	1 0	0 0	Light Heavyweight	CNC	1	UFC 294: Makhachev vs. Volkanovski 2
3	Ikram Aliskerov Warlley Alves	1 0	26 12	0 0	0 0	Middleweight	KO/TKO Punches	1	UFC 294: Makhachev vs. Volkanovski 2
4	Said Nurmagomedov Muin Gafurov	0 0	5 1	0 0	1 0	Bantamweight	SUB Guillotine Choke	1	UFC 294: Makhachev vs. Volkanovski 2

Now, I will convert the strings in "Kd", "Str", "Td", and "Sub" into lists of numbers. I will change non-numeric items (in this case "--") into 0's.

def convert_to_int(x):
    return [int(i) if isinstance(i, str) and i != '--' else 0 for i in x] if isinstance(x, list) else [0]

data_1["Kd"] = data_1["Kd"].str.split().apply(convert_to_int)
data_1["Str"] = data_1["Str"].str.split().apply(convert_to_int)
data_1["Td"] = data_1["Td"].str.split().apply(convert_to_int)
data_1["Sub"] = data_1["Sub"].str.split().apply(convert_to_int)

data_1.head()

	Fighter	Kd	Str	Td	Sub	Weight class	Method	Round	Event
0	Islam Makhachev Alexander Volkanovski	[1, 0]	[24, 4]	[0, 0]	[0, 0]	Lightweight	KO/TKO Kick	1	UFC 294: Makhachev vs. Volkanovski 2
1	Khamzat Chimaev Kamaru Usman	[0, 0]	[38, 36]	[4, 0]	[1, 0]	Middleweight	M-DEC	3	UFC 294: Makhachev vs. Volkanovski 2
2	Magomed Ankalaev Johnny Walker	[0, 0]	[13, 14]	[1, 0]	[0, 0]	Light Heavyweight	CNC	1	UFC 294: Makhachev vs. Volkanovski 2
3	Ikram Aliskerov Warlley Alves	[1, 0]	[26, 12]	[0, 0]	[0, 0]	Middleweight	KO/TKO Punches	1	UFC 294: Makhachev vs. Volkanovski 2
4	Said Nurmagomedov Muin Gafurov	[0, 0]	[5, 1]	[0, 0]	[1, 0]	Bantamweight	SUB Guillotine Choke	1	UFC 294: Makhachev vs. Volkanovski 2

The data types are displayed below. Further cleaning will be done after the datasets are merged.

data_1.dtypes

Fighter         object
Kd              object
Str             object
Td              object
Sub             object
Weight class    object
Method          object
Round            int64
Event           object
dtype: object

I will now import the second dataset. I remove unnecessary columns and remove leading and trailing whitespace from the "Fighter" column. There are duplicate rows in this dataset, which I also remove.

# Loading the second dataset
data_2 = pd.read_csv("ufc_compiled.csv")
data_2.head()

	Unnamed: 0	Fighter	Head	Body	Leg	Distance	Clinch	Ground	Event
0	0	Islam Makhachev Alexander Volkanovski	13 of 16 0 of 1	9 of 11 1 of 1	2 of 2 3 of 4	7 of 11 3 of 5	8 of 9 1 of 1	9 of 9 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
1	1	Kamaru Usman Khamzat Chimaev	27 of 52 27 of 53	6 of 10 3 of 9	3 of 4 8 of 8	33 of 63 22 of 53	1 of 1 0 of 0	2 of 2 16 of 17	UFC 294: Makhachev vs. Volkanovski 2
2	2	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
3	3	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2
4	4	Ikram Aliskerov Warlley Alves	20 of 29 3 of 6	6 of 6 1 of 1	0 of 1 8 of 9	26 of 36 12 of 16	0 of 0 0 of 0	0 of 0 0 of 0	UFC 294: Makhachev vs. Volkanovski 2

# Loading the third dataset
data_3 = pd.read_csv("ufc_third.csv")
data_3.head()

	Unnamed: 0	Fighter	Sig. str.	Td	Rev.	Ctrl	Event
0	0	Islam Makhachev Alexander Volkanovski	24 of 29 4 of 6	0 of 1 0 of 0	0 0	0:19 0:49	UFC 294: Makhachev vs. Volkanovski 2
1	1	Kamaru Usman Khamzat Chimaev	36 of 66 38 of 70	0 of 1 4 of 12	0 0	0:04 7:16	UFC 294: Makhachev vs. Volkanovski 2
2	2	Magomed Ankalaev Johnny Walker	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00	UFC 294: Makhachev vs. Volkanovski 2
3	3	Magomed Ankalaev Johnny Walker	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00	UFC 294: Makhachev vs. Volkanovski 2
4	4	Ikram Aliskerov Warlley Alves	26 of 36 12 of 16	0 of 0 0 of 0	0 0	0:00 0:00	UFC 294: Makhachev vs. Volkanovski 2

# Merging the datasets
data_adj = pd.merge(data_2, data_3, on=['Fighter', "Event"], how='inner')
data_adj.head()

	Unnamed: 0_x	Fighter	Head	Body	Leg	Distance	Clinch	Ground	Event	Unnamed: 0_y	Sig. str.	Td	Rev.	Ctrl
0	0	Islam Makhachev Alexander Volkanovski	13 of 16 0 of 1	9 of 11 1 of 1	2 of 2 3 of 4	7 of 11 3 of 5	8 of 9 1 of 1	9 of 9 0 of 0	UFC 294: Makhachev vs. Volkanovski 2	0	24 of 29 4 of 6	0 of 1 0 of 0	0 0	0:19 0:49
1	1	Kamaru Usman Khamzat Chimaev	27 of 52 27 of 53	6 of 10 3 of 9	3 of 4 8 of 8	33 of 63 22 of 53	1 of 1 0 of 0	2 of 2 16 of 17	UFC 294: Makhachev vs. Volkanovski 2	1	36 of 66 38 of 70	0 of 1 4 of 12	0 0	0:04 7:16
2	2	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2	2	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00
3	2	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2	3	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00
4	3	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2	2	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00

# Renaming columns
data_adj.rename(columns={"Fighter": "Fighter_1"}, inplace=True)
data_adj["Fighter_1"] = data_adj["Fighter_1"].str.strip()
data_adj.head()

	Unnamed: 0_x	Fighter_1	Head	Body	Leg	Distance	Clinch	Ground	Event	Unnamed: 0_y	Sig. str.	Td	Rev.	Ctrl
0	0	Islam Makhachev Alexander Volkanovski	13 of 16 0 of 1	9 of 11 1 of 1	2 of 2 3 of 4	7 of 11 3 of 5	8 of 9 1 of 1	9 of 9 0 of 0	UFC 294: Makhachev vs. Volkanovski 2	0	24 of 29 4 of 6	0 of 1 0 of 0	0 0	0:19 0:49
1	1	Kamaru Usman Khamzat Chimaev	27 of 52 27 of 53	6 of 10 3 of 9	3 of 4 8 of 8	33 of 63 22 of 53	1 of 1 0 of 0	2 of 2 16 of 17	UFC 294: Makhachev vs. Volkanovski 2	1	36 of 66 38 of 70	0 of 1 4 of 12	0 0	0:04 7:16
2	2	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2	2	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00
3	2	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2	3	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00
4	3	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2	2	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00

# Removing duplicates
data_adj = data_adj.drop_duplicates(subset=["Fighter_1"])
data_adj = data_adj.reset_index(drop=True)
data_adj.head()

	Unnamed: 0_x	Fighter_1	Head	Body	Leg	Distance	Clinch	Ground	Event	Unnamed: 0_y	Sig. str.	Td	Rev.	Ctrl
0	0	Islam Makhachev Alexander Volkanovski	13 of 16 0 of 1	9 of 11 1 of 1	2 of 2 3 of 4	7 of 11 3 of 5	8 of 9 1 of 1	9 of 9 0 of 0	UFC 294: Makhachev vs. Volkanovski 2	0	24 of 29 4 of 6	0 of 1 0 of 0	0 0	0:19 0:49
1	1	Kamaru Usman Khamzat Chimaev	27 of 52 27 of 53	6 of 10 3 of 9	3 of 4 8 of 8	33 of 63 22 of 53	1 of 1 0 of 0	2 of 2 16 of 17	UFC 294: Makhachev vs. Volkanovski 2	1	36 of 66 38 of 70	0 of 1 4 of 12	0 0	0:04 7:16
2	2	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	UFC 294: Makhachev vs. Volkanovski 2	2	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00
3	4	Ikram Aliskerov Warlley Alves	20 of 29 3 of 6	6 of 6 1 of 1	0 of 1 8 of 9	26 of 36 12 of 16	0 of 0 0 of 0	0 of 0 0 of 0	UFC 294: Makhachev vs. Volkanovski 2	4	26 of 36 12 of 16	0 of 0 0 of 0	0 0	0:00 0:00
4	5	Said Nurmagomedov Muin Gafurov	2 of 7 0 of 2	2 of 2 0 of 0	1 of 2 1 of 1	5 of 11 1 of 3	0 of 0 0 of 0	0 of 0 0 of 0	UFC 294: Makhachev vs. Volkanovski 2	5	5 of 11 1 of 3	0 of 0 0 of 1	0 0	0:03 0:07

I now join the dataframes on the "Fighter" and "Event" columns.

# Joining the datasets
data_1['key'] = data_1['Fighter'].str.split().apply(sorted).str.join(' ')
data_adj['key'] = data_adj['Fighter_1'].str.split().apply(sorted).str.join(' ')

merged_df = data_1.merge(data_adj, on=['key', 'Event'], how='outer')

merged_df = merged_df.drop(columns=['key'])

merged_df.head()

	Fighter	Kd	Str	Td_x	Sub	Weight class	Method	Round	Event	Unnamed: 0_x	...	Body	Leg	Distance	Clinch	Ground	Unnamed: 0_y	Sig. str.	Td_y	Rev.	Ctrl
0	Islam Makhachev Alexander Volkanovski	[1, 0]	[24, 4]	[0, 0]	[0, 0]	Lightweight	KO/TKO Kick	1	UFC 294: Makhachev vs. Volkanovski 2	0.0	...	9 of 11 1 of 1	2 of 2 3 of 4	7 of 11 3 of 5	8 of 9 1 of 1	9 of 9 0 of 0	0.0	24 of 29 4 of 6	0 of 1 0 of 0	0 0	0:19 0:49
1	Khamzat Chimaev Kamaru Usman	[0, 0]	[38, 36]	[4, 0]	[1, 0]	Middleweight	M-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	1.0	...	6 of 10 3 of 9	3 of 4 8 of 8	33 of 63 22 of 53	1 of 1 0 of 0	2 of 2 16 of 17	1.0	36 of 66 38 of 70	0 of 1 4 of 12	0 0	0:04 7:16
2	Magomed Ankalaev Johnny Walker	[0, 0]	[13, 14]	[1, 0]	[0, 0]	Light Heavyweight	CNC	1	UFC 294: Makhachev vs. Volkanovski 2	2.0	...	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	2.0	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00
3	Ikram Aliskerov Warlley Alves	[1, 0]	[26, 12]	[0, 0]	[0, 0]	Middleweight	KO/TKO Punches	1	UFC 294: Makhachev vs. Volkanovski 2	4.0	...	6 of 6 1 of 1	0 of 1 8 of 9	26 of 36 12 of 16	0 of 0 0 of 0	0 of 0 0 of 0	4.0	26 of 36 12 of 16	0 of 0 0 of 0	0 0	0:00 0:00
4	Said Nurmagomedov Muin Gafurov	[0, 0]	[5, 1]	[0, 0]	[1, 0]	Bantamweight	SUB Guillotine Choke	1	UFC 294: Makhachev vs. Volkanovski 2	5.0	...	2 of 2 0 of 0	1 of 2 1 of 1	5 of 11 1 of 3	0 of 0 0 of 0	0 of 0 0 of 0	5.0	5 of 11 1 of 3	0 of 0 0 of 1	0 0	0:03 0:07

5 rows × 22 columns

Some fight data is unavailable, and so fights with unavailable data will be dropped.

# Removing unavailable data
merged_df = merged_df.dropna()
merged_df.head()

	Fighter	Kd	Str	Td_x	Sub	Weight class	Method	Round	Event	Unnamed: 0_x	...	Body	Leg	Distance	Clinch	Ground	Unnamed: 0_y	Sig. str.	Td_y	Rev.	Ctrl
0	Islam Makhachev Alexander Volkanovski	[1, 0]	[24, 4]	[0, 0]	[0, 0]	Lightweight	KO/TKO Kick	1	UFC 294: Makhachev vs. Volkanovski 2	0.0	...	9 of 11 1 of 1	2 of 2 3 of 4	7 of 11 3 of 5	8 of 9 1 of 1	9 of 9 0 of 0	0.0	24 of 29 4 of 6	0 of 1 0 of 0	0 0	0:19 0:49
1	Khamzat Chimaev Kamaru Usman	[0, 0]	[38, 36]	[4, 0]	[1, 0]	Middleweight	M-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	1.0	...	6 of 10 3 of 9	3 of 4 8 of 8	33 of 63 22 of 53	1 of 1 0 of 0	2 of 2 16 of 17	1.0	36 of 66 38 of 70	0 of 1 4 of 12	0 0	0:04 7:16
2	Magomed Ankalaev Johnny Walker	[0, 0]	[13, 14]	[1, 0]	[0, 0]	Light Heavyweight	CNC	1	UFC 294: Makhachev vs. Volkanovski 2	2.0	...	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	2.0	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00
3	Ikram Aliskerov Warlley Alves	[1, 0]	[26, 12]	[0, 0]	[0, 0]	Middleweight	KO/TKO Punches	1	UFC 294: Makhachev vs. Volkanovski 2	4.0	...	6 of 6 1 of 1	0 of 1 8 of 9	26 of 36 12 of 16	0 of 0 0 of 0	0 of 0 0 of 0	4.0	26 of 36 12 of 16	0 of 0 0 of 0	0 0	0:00 0:00
4	Said Nurmagomedov Muin Gafurov	[0, 0]	[5, 1]	[0, 0]	[1, 0]	Bantamweight	SUB Guillotine Choke	1	UFC 294: Makhachev vs. Volkanovski 2	5.0	...	2 of 2 0 of 0	1 of 2 1 of 1	5 of 11 1 of 3	0 of 0 0 of 0	0 of 0 0 of 0	5.0	5 of 11 1 of 3	0 of 0 0 of 1	0 0	0:03 0:07

5 rows × 22 columns

Before further analysis, I drop superfluous columns.

# Dropping unnecessary columns
merged_df = merged_df.drop(["Unnamed: 0_x", "Unnamed: 0_y"], axis=1)
merged_df.head()

	Fighter	Kd	Str	Td_x	Sub	Weight class	Method	Round	Event	Fighter_1	Head	Body	Leg	Distance	Clinch	Ground	Sig. str.	Td_y	Rev.	Ctrl
0	Islam Makhachev Alexander Volkanovski	[1, 0]	[24, 4]	[0, 0]	[0, 0]	Lightweight	KO/TKO Kick	1	UFC 294: Makhachev vs. Volkanovski 2	Islam Makhachev Alexander Volkanovski	13 of 16 0 of 1	9 of 11 1 of 1	2 of 2 3 of 4	7 of 11 3 of 5	8 of 9 1 of 1	9 of 9 0 of 0	24 of 29 4 of 6	0 of 1 0 of 0	0 0	0:19 0:49
1	Khamzat Chimaev Kamaru Usman	[0, 0]	[38, 36]	[4, 0]	[1, 0]	Middleweight	M-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Kamaru Usman Khamzat Chimaev	27 of 52 27 of 53	6 of 10 3 of 9	3 of 4 8 of 8	33 of 63 22 of 53	1 of 1 0 of 0	2 of 2 16 of 17	36 of 66 38 of 70	0 of 1 4 of 12	0 0	0:04 7:16
2	Magomed Ankalaev Johnny Walker	[0, 0]	[13, 14]	[1, 0]	[0, 0]	Light Heavyweight	CNC	1	UFC 294: Makhachev vs. Volkanovski 2	Magomed Ankalaev Johnny Walker	7 of 13 4 of 15	3 of 3 1 of 2	3 of 3 9 of 10	7 of 11 14 of 27	1 of 1 0 of 0	5 of 7 0 of 0	13 of 19 14 of 27	1 of 1 0 of 0	0 0	1:06 0:00
3	Ikram Aliskerov Warlley Alves	[1, 0]	[26, 12]	[0, 0]	[0, 0]	Middleweight	KO/TKO Punches	1	UFC 294: Makhachev vs. Volkanovski 2	Ikram Aliskerov Warlley Alves	20 of 29 3 of 6	6 of 6 1 of 1	0 of 1 8 of 9	26 of 36 12 of 16	0 of 0 0 of 0	0 of 0 0 of 0	26 of 36 12 of 16	0 of 0 0 of 0	0 0	0:00 0:00
4	Said Nurmagomedov Muin Gafurov	[0, 0]	[5, 1]	[0, 0]	[1, 0]	Bantamweight	SUB Guillotine Choke	1	UFC 294: Makhachev vs. Volkanovski 2	Said Nurmagomedov Muin Gafurov	2 of 7 0 of 2	2 of 2 0 of 0	1 of 2 1 of 1	5 of 11 1 of 3	0 of 0 0 of 0	0 of 0 0 of 0	5 of 11 1 of 3	0 of 0 0 of 1	0 0	0:03 0:07

Now, I extract all data that ended in a decision that went 3 or 5 rounds. Some early UFC decisions took place in fights that only went 1 or 2 rounds, and as all fights that take place today go 3 or 5 rounds, these old fights do not apply to future predictions.

decision_data = merged_df[merged_df['Method'].str.contains('DEC', case=False)].reset_index(drop=True)
decision_data = decision_data[(decision_data["Round"] == 3) | (decision_data["Round"] == 5)]
decision_data.head()

	Fighter	Kd	Str	Td_x	Sub	Weight class	Method	Round	Event	Fighter_1	Head	Body	Leg	Distance	Clinch	Ground	Sig. str.	Td_y	Rev.	Ctrl
0	Khamzat Chimaev Kamaru Usman	[0, 0]	[38, 36]	[4, 0]	[1, 0]	Middleweight	M-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Kamaru Usman Khamzat Chimaev	27 of 52 27 of 53	6 of 10 3 of 9	3 of 4 8 of 8	33 of 63 22 of 53	1 of 1 0 of 0	2 of 2 16 of 17	36 of 66 38 of 70	0 of 1 4 of 12	0 0	0:04 7:16
1	Trevor Peek Mohammad Yahya	[0, 0]	[61, 43]	[4, 1]	[0, 1]	Lightweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Mohammad Yahya Trevor Peek	28 of 70 32 of 61	14 of 21 16 of 23	1 of 1 13 of 13	41 of 90 50 of 83	2 of 2 6 of 9	0 of 0 5 of 5	43 of 92 61 of 97	1 of 1 4 of 7	0 0	0:24 5:10
2	Sedriques Dumas Abu Azaitar	[0, 0]	[34, 41]	[1, 2]	[0, 0]	Middleweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Abu Azaitar Sedriques Dumas	21 of 78 11 of 45	13 of 18 5 of 11	7 of 10 18 of 23	31 of 90 31 of 76	3 of 6 0 of 0	7 of 10 3 of 3	41 of 106 34 of 79	2 of 4 1 of 2	0 0	0:32 4:10
3	Muhammad Naimov Nathaniel Wood	[0, 0]	[48, 50]	[4, 2]	[0, 0]	Featherweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Nathaniel Wood Muhammad Naimov	37 of 57 31 of 60	6 of 7 10 of 13	7 of 8 7 of 8	36 of 56 24 of 48	5 of 6 11 of 14	9 of 10 13 of 19	50 of 72 48 of 81	2 of 7 4 of 9	0 1	2:44 6:42
4	Viktoriia Dudakova Jinh Yu Frey	[1, 0]	[74, 47]	[0, 1]	[0, 0]	Women's Strawweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Viktoriia Dudakova Jinh Yu Frey	47 of 115 31 of 80	26 of 30 8 of 13	1 of 1 8 of 8	60 of 131 37 of 89	14 of 15 3 of 4	0 of 0 7 of 8	74 of 146 47 of 101	0 of 3 1 of 3	0 0	0:18 4:40

Because both fighters' data are in the same column for "Head", "Body", "Leg", "Distance", "Clinch", and "Ground". I need to put the data in separate columns.

def string_to_numbers(column):
    numbers = re.findall(r'\d+', str(column))  
    return [int(num) for num in numbers]

decision_data["Head"] = decision_data["Head"].apply(string_to_numbers)
decision_data["Body"] = decision_data["Body"].apply(string_to_numbers)
decision_data["Leg"] = decision_data["Leg"].apply(string_to_numbers)
decision_data["Distance"] = decision_data["Distance"].apply(string_to_numbers)
decision_data["Clinch"] = decision_data["Clinch"].apply(string_to_numbers)
decision_data["Ground"] = decision_data["Ground"].apply(string_to_numbers)
decision_data["Sig. str."] = decision_data["Sig. str."].apply(string_to_numbers)
decision_data["Td_y"] = decision_data["Td_y"].apply(string_to_numbers)
decision_data["Rev."] = decision_data["Rev."].apply(string_to_numbers)

decision_data.head()

	Fighter	Kd	Str	Td_x	Sub	Weight class	Method	Round	Event	Fighter_1	Head	Body	Leg	Distance	Clinch	Ground	Sig. str.	Td_y	Rev.	Ctrl
0	Khamzat Chimaev Kamaru Usman	[0, 0]	[38, 36]	[4, 0]	[1, 0]	Middleweight	M-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Kamaru Usman Khamzat Chimaev	[27, 52, 27, 53]	[6, 10, 3, 9]	[3, 4, 8, 8]	[33, 63, 22, 53]	[1, 1, 0, 0]	[2, 2, 16, 17]	[36, 66, 38, 70]	[0, 1, 4, 12]	[0, 0]	0:04 7:16
1	Trevor Peek Mohammad Yahya	[0, 0]	[61, 43]	[4, 1]	[0, 1]	Lightweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Mohammad Yahya Trevor Peek	[28, 70, 32, 61]	[14, 21, 16, 23]	[1, 1, 13, 13]	[41, 90, 50, 83]	[2, 2, 6, 9]	[0, 0, 5, 5]	[43, 92, 61, 97]	[1, 1, 4, 7]	[0, 0]	0:24 5:10
2	Sedriques Dumas Abu Azaitar	[0, 0]	[34, 41]	[1, 2]	[0, 0]	Middleweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Abu Azaitar Sedriques Dumas	[21, 78, 11, 45]	[13, 18, 5, 11]	[7, 10, 18, 23]	[31, 90, 31, 76]	[3, 6, 0, 0]	[7, 10, 3, 3]	[41, 106, 34, 79]	[2, 4, 1, 2]	[0, 0]	0:32 4:10
3	Muhammad Naimov Nathaniel Wood	[0, 0]	[48, 50]	[4, 2]	[0, 0]	Featherweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Nathaniel Wood Muhammad Naimov	[37, 57, 31, 60]	[6, 7, 10, 13]	[7, 8, 7, 8]	[36, 56, 24, 48]	[5, 6, 11, 14]	[9, 10, 13, 19]	[50, 72, 48, 81]	[2, 7, 4, 9]	[0, 1]	2:44 6:42
4	Viktoriia Dudakova Jinh Yu Frey	[1, 0]	[74, 47]	[0, 1]	[0, 0]	Women's Strawweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Viktoriia Dudakova Jinh Yu Frey	[47, 115, 31, 80]	[26, 30, 8, 13]	[1, 1, 8, 8]	[60, 131, 37, 89]	[14, 15, 3, 4]	[0, 0, 7, 8]	[74, 146, 47, 101]	[0, 3, 1, 3]	[0, 0]	0:18 4:40

This part of the data cleaning is difficult. The fighter columns do not have the fighters in the same order. Because the order of fighters is random in the second "Fighter" column, I will use this column to determine "Fighter_1" and "Fighter_2" data regarding. If the columns are the same, then "Fighter_1" will take on the first number in each column. If not, then "Fighter_1" will take on the second value.

decision_data['Fighter_1_Kd'] = decision_data.apply(lambda row: row['Kd'][0] if row['Fighter_1'] == row['Fighter'] else row['Kd'][1], axis=1)
decision_data['Fighter_2_Kd'] = decision_data.apply(lambda row: row['Kd'][1] if row['Fighter_1'] == row['Fighter'] else row['Kd'][0], axis=1)
decision_data['Fighter_1_Str'] = decision_data.apply(lambda row: row['Str'][0] if row['Fighter_1'] == row['Fighter'] else row['Str'][1], axis=1)
decision_data['Fighter_2_Str'] = decision_data.apply(lambda row: row['Str'][1] if row['Fighter_1'] == row['Fighter'] else row['Str'][0], axis=1)
decision_data['Fighter_1_Td'] = decision_data.apply(lambda row: row['Td_x'][0] if row['Fighter_1'] == row['Fighter'] else row['Td_x'][1], axis=1)
decision_data['Fighter_2_Td'] = decision_data.apply(lambda row: row['Td_x'][1] if row['Fighter_1'] == row['Fighter'] else row['Td_x'][0], axis=1)
decision_data['Fighter_1_Sub'] = decision_data.apply(lambda row: row['Sub'][0] if row['Fighter_1'] == row['Fighter'] else row['Sub'][1], axis=1)
decision_data['Fighter_2_Sub'] = decision_data.apply(lambda row: row['Sub'][1] if row['Fighter_1'] == row['Fighter'] else row['Sub'][0], axis=1)
decision_data.head()

	Fighter	Kd	Str	Td_x	Sub	Weight class	Method	Round	Event	Fighter_1	...	Rev.	Ctrl	Fighter_1_Kd	Fighter_2_Kd	Fighter_1_Str	Fighter_2_Str	Fighter_1_Td	Fighter_2_Td	Fighter_1_Sub	Fighter_2_Sub
0	Khamzat Chimaev Kamaru Usman	[0, 0]	[38, 36]	[4, 0]	[1, 0]	Middleweight	M-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Kamaru Usman Khamzat Chimaev	...	[0, 0]	0:04 7:16	0	0	36	38	0	4	0	1
1	Trevor Peek Mohammad Yahya	[0, 0]	[61, 43]	[4, 1]	[0, 1]	Lightweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Mohammad Yahya Trevor Peek	...	[0, 0]	0:24 5:10	0	0	43	61	1	4	1	0
2	Sedriques Dumas Abu Azaitar	[0, 0]	[34, 41]	[1, 2]	[0, 0]	Middleweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Abu Azaitar Sedriques Dumas	...	[0, 0]	0:32 4:10	0	0	41	34	2	1	0	0
3	Muhammad Naimov Nathaniel Wood	[0, 0]	[48, 50]	[4, 2]	[0, 0]	Featherweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Nathaniel Wood Muhammad Naimov	...	[0, 1]	2:44 6:42	0	0	50	48	2	4	0	0
4	Viktoriia Dudakova Jinh Yu Frey	[1, 0]	[74, 47]	[0, 1]	[0, 0]	Women's Strawweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Viktoriia Dudakova Jinh Yu Frey	...	[0, 0]	0:18 4:40	0	1	47	74	1	0	0	0

5 rows × 28 columns

Now, I remove the columns from which the data was extracted.

decision_data = decision_data.drop(columns=["Kd", "Str", "Td_x", "Sub"], axis=1)

Now, I extract data from the "Head", "Body", "Leg", "Distance", "Clinch", and "Ground" columns.

def extract_values_from_list(lst):
    if len(lst) == 4:
        return lst[0], lst[2]
    else:
        return None, None

decision_data[["Fighter_1_Head", "Fighter_2_Head"]] = decision_data["Head"].apply(lambda x: pd.Series(extract_values_from_list(x)))
decision_data[["Fighter_1_Body", "Fighter_2_Body"]] = decision_data["Body"].apply(lambda x: pd.Series(extract_values_from_list(x)))
decision_data[["Fighter_1_Leg", "Fighter_2_Leg"]] = decision_data["Leg"].apply(lambda x: pd.Series(extract_values_from_list(x)))
decision_data[["Fighter_1_Distance", "Fighter_2_Distance"]] = decision_data["Distance"].apply(lambda x: pd.Series(extract_values_from_list(x)))
decision_data[["Fighter_1_Clinch", "Fighter_2_Clinch"]] = decision_data["Clinch"].apply(lambda x: pd.Series(extract_values_from_list(x)))
decision_data[["Fighter_1_Ground", "Fighter_2_Ground"]] = decision_data["Ground"].apply(lambda x: pd.Series(extract_values_from_list(x)))

decision_data.head()

	Fighter	Weight class	Method	Round	Event	Fighter_1	Head	Body	Leg	Distance	...	Fighter_1_Body	Fighter_2_Body	Fighter_1_Leg	Fighter_2_Leg	Fighter_1_Distance	Fighter_2_Distance	Fighter_1_Clinch	Fighter_2_Clinch	Fighter_1_Ground	Fighter_2_Ground
0	Khamzat Chimaev Kamaru Usman	Middleweight	M-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Kamaru Usman Khamzat Chimaev	[27, 52, 27, 53]	[6, 10, 3, 9]	[3, 4, 8, 8]	[33, 63, 22, 53]	...	6	3	3	8	33	22	1	0	2	16
1	Trevor Peek Mohammad Yahya	Lightweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Mohammad Yahya Trevor Peek	[28, 70, 32, 61]	[14, 21, 16, 23]	[1, 1, 13, 13]	[41, 90, 50, 83]	...	14	16	1	13	41	50	2	6	0	5
2	Sedriques Dumas Abu Azaitar	Middleweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Abu Azaitar Sedriques Dumas	[21, 78, 11, 45]	[13, 18, 5, 11]	[7, 10, 18, 23]	[31, 90, 31, 76]	...	13	5	7	18	31	31	3	0	7	3
3	Muhammad Naimov Nathaniel Wood	Featherweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Nathaniel Wood Muhammad Naimov	[37, 57, 31, 60]	[6, 7, 10, 13]	[7, 8, 7, 8]	[36, 56, 24, 48]	...	6	10	7	7	36	24	5	11	9	13
4	Viktoriia Dudakova Jinh Yu Frey	Women's Strawweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Viktoriia Dudakova Jinh Yu Frey	[47, 115, 31, 80]	[26, 30, 8, 13]	[1, 1, 8, 8]	[60, 131, 37, 89]	...	26	8	1	8	60	37	14	3	0	7

5 rows × 36 columns

# Extracting Takedown Attempts

def extract_values_from_list_1(lst):
    if len(lst) == 4:
        return lst[1], lst[3]
    else:
        return None, None
    
decision_data[["Fighter_1_Td_Att", "Fighter_2_Td_Att"]] = decision_data["Td_y"].apply(lambda x: pd.Series(extract_values_from_list_1(x)))
decision_data.head()

	Fighter	Weight class	Method	Round	Event	Fighter_1	Head	Body	Leg	Distance	...	Fighter_1_Leg	Fighter_2_Leg	Fighter_1_Distance	Fighter_2_Distance	Fighter_1_Clinch	Fighter_2_Clinch	Fighter_1_Ground	Fighter_2_Ground	Fighter_1_Td_Att	Fighter_2_Td_Att
0	Khamzat Chimaev Kamaru Usman	Middleweight	M-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Kamaru Usman Khamzat Chimaev	[27, 52, 27, 53]	[6, 10, 3, 9]	[3, 4, 8, 8]	[33, 63, 22, 53]	...	3	8	33	22	1	0	2	16	1	12
1	Trevor Peek Mohammad Yahya	Lightweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Mohammad Yahya Trevor Peek	[28, 70, 32, 61]	[14, 21, 16, 23]	[1, 1, 13, 13]	[41, 90, 50, 83]	...	1	13	41	50	2	6	0	5	1	7
2	Sedriques Dumas Abu Azaitar	Middleweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Abu Azaitar Sedriques Dumas	[21, 78, 11, 45]	[13, 18, 5, 11]	[7, 10, 18, 23]	[31, 90, 31, 76]	...	7	18	31	31	3	0	7	3	4	2
3	Muhammad Naimov Nathaniel Wood	Featherweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Nathaniel Wood Muhammad Naimov	[37, 57, 31, 60]	[6, 7, 10, 13]	[7, 8, 7, 8]	[36, 56, 24, 48]	...	7	7	36	24	5	11	9	13	7	9
4	Viktoriia Dudakova Jinh Yu Frey	Women's Strawweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Viktoriia Dudakova Jinh Yu Frey	[47, 115, 31, 80]	[26, 30, 8, 13]	[1, 1, 8, 8]	[60, 131, 37, 89]	...	1	8	60	37	14	3	0	7	3	3

5 rows × 38 columns

# Adjusting the Control Time Column

decision_data["Ctrl"] = decision_data["Ctrl"].str.split()
decision_data.head()

	Fighter	Weight class	Method	Round	Event	Fighter_1	Head	Body	Leg	Distance	...	Fighter_1_Leg	Fighter_2_Leg	Fighter_1_Distance	Fighter_2_Distance	Fighter_1_Clinch	Fighter_2_Clinch	Fighter_1_Ground	Fighter_2_Ground	Fighter_1_Td_Att	Fighter_2_Td_Att
0	Khamzat Chimaev Kamaru Usman	Middleweight	M-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Kamaru Usman Khamzat Chimaev	[27, 52, 27, 53]	[6, 10, 3, 9]	[3, 4, 8, 8]	[33, 63, 22, 53]	...	3	8	33	22	1	0	2	16	1	12
1	Trevor Peek Mohammad Yahya	Lightweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Mohammad Yahya Trevor Peek	[28, 70, 32, 61]	[14, 21, 16, 23]	[1, 1, 13, 13]	[41, 90, 50, 83]	...	1	13	41	50	2	6	0	5	1	7
2	Sedriques Dumas Abu Azaitar	Middleweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Abu Azaitar Sedriques Dumas	[21, 78, 11, 45]	[13, 18, 5, 11]	[7, 10, 18, 23]	[31, 90, 31, 76]	...	7	18	31	31	3	0	7	3	4	2
3	Muhammad Naimov Nathaniel Wood	Featherweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Nathaniel Wood Muhammad Naimov	[37, 57, 31, 60]	[6, 7, 10, 13]	[7, 8, 7, 8]	[36, 56, 24, 48]	...	7	7	36	24	5	11	9	13	7	9
4	Viktoriia Dudakova Jinh Yu Frey	Women's Strawweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Viktoriia Dudakova Jinh Yu Frey	[47, 115, 31, 80]	[26, 30, 8, 13]	[1, 1, 8, 8]	[60, 131, 37, 89]	...	1	8	60	37	14	3	0	7	3	3

5 rows × 38 columns

# Extracting Control Times

def extract_values_from_list_2(lst):
    if len(lst) == 2:
        return lst[0], lst[1]
    else:
        return None, None
    
decision_data[["Fighter_1_Ctrl", "Fighter_2_Ctrl"]] = decision_data["Ctrl"].apply(lambda x: pd.Series(extract_values_from_list_2(x)))
decision_data.head()

	Fighter	Weight class	Method	Round	Event	Fighter_1	Head	Body	Leg	Distance	...	Fighter_1_Distance	Fighter_2_Distance	Fighter_1_Clinch	Fighter_2_Clinch	Fighter_1_Ground	Fighter_2_Ground	Fighter_1_Td_Att	Fighter_2_Td_Att	Fighter_1_Ctrl	Fighter_2_Ctrl
0	Khamzat Chimaev Kamaru Usman	Middleweight	M-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Kamaru Usman Khamzat Chimaev	[27, 52, 27, 53]	[6, 10, 3, 9]	[3, 4, 8, 8]	[33, 63, 22, 53]	...	33	22	1	0	2	16	1	12	0:04	7:16
1	Trevor Peek Mohammad Yahya	Lightweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Mohammad Yahya Trevor Peek	[28, 70, 32, 61]	[14, 21, 16, 23]	[1, 1, 13, 13]	[41, 90, 50, 83]	...	41	50	2	6	0	5	1	7	0:24	5:10
2	Sedriques Dumas Abu Azaitar	Middleweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Abu Azaitar Sedriques Dumas	[21, 78, 11, 45]	[13, 18, 5, 11]	[7, 10, 18, 23]	[31, 90, 31, 76]	...	31	31	3	0	7	3	4	2	0:32	4:10
3	Muhammad Naimov Nathaniel Wood	Featherweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Nathaniel Wood Muhammad Naimov	[37, 57, 31, 60]	[6, 7, 10, 13]	[7, 8, 7, 8]	[36, 56, 24, 48]	...	36	24	5	11	9	13	7	9	2:44	6:42
4	Viktoriia Dudakova Jinh Yu Frey	Women's Strawweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Viktoriia Dudakova Jinh Yu Frey	[47, 115, 31, 80]	[26, 30, 8, 13]	[1, 1, 8, 8]	[60, 131, 37, 89]	...	60	37	14	3	0	7	3	3	0:18	4:40

5 rows × 40 columns

# Changing missing times

decision_data = decision_data.replace('--', '00:00')

decision_data.head()

	Fighter	Weight class	Method	Round	Event	Fighter_1	Head	Body	Leg	Distance	...	Fighter_1_Distance	Fighter_2_Distance	Fighter_1_Clinch	Fighter_2_Clinch	Fighter_1_Ground	Fighter_2_Ground	Fighter_1_Td_Att	Fighter_2_Td_Att	Fighter_1_Ctrl	Fighter_2_Ctrl
0	Khamzat Chimaev Kamaru Usman	Middleweight	M-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Kamaru Usman Khamzat Chimaev	[27, 52, 27, 53]	[6, 10, 3, 9]	[3, 4, 8, 8]	[33, 63, 22, 53]	...	33	22	1	0	2	16	1	12	0:04	7:16
1	Trevor Peek Mohammad Yahya	Lightweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Mohammad Yahya Trevor Peek	[28, 70, 32, 61]	[14, 21, 16, 23]	[1, 1, 13, 13]	[41, 90, 50, 83]	...	41	50	2	6	0	5	1	7	0:24	5:10
2	Sedriques Dumas Abu Azaitar	Middleweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Abu Azaitar Sedriques Dumas	[21, 78, 11, 45]	[13, 18, 5, 11]	[7, 10, 18, 23]	[31, 90, 31, 76]	...	31	31	3	0	7	3	4	2	0:32	4:10
3	Muhammad Naimov Nathaniel Wood	Featherweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Nathaniel Wood Muhammad Naimov	[37, 57, 31, 60]	[6, 7, 10, 13]	[7, 8, 7, 8]	[36, 56, 24, 48]	...	36	24	5	11	9	13	7	9	2:44	6:42
4	Viktoriia Dudakova Jinh Yu Frey	Women's Strawweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Viktoriia Dudakova Jinh Yu Frey	[47, 115, 31, 80]	[26, 30, 8, 13]	[1, 1, 8, 8]	[60, 131, 37, 89]	...	60	37	14	3	0	7	3	3	0:18	4:40

5 rows × 40 columns

# Continuing to adjust the control time columns

decision_data['Fighter_1_Ctrl'] = decision_data['Fighter_1_Ctrl'].apply(lambda x: "00:0" + x if len(x) == 4 else "00:" + x)
decision_data['Fighter_2_Ctrl'] = decision_data['Fighter_2_Ctrl'].apply(lambda x: "00:0" + x if len(x) == 4 else "00:" + x)

decision_data.head()

	Fighter	Weight class	Method	Round	Event	Fighter_1	Head	Body	Leg	Distance	...	Fighter_1_Distance	Fighter_2_Distance	Fighter_1_Clinch	Fighter_2_Clinch	Fighter_1_Ground	Fighter_2_Ground	Fighter_1_Td_Att	Fighter_2_Td_Att	Fighter_1_Ctrl	Fighter_2_Ctrl
0	Khamzat Chimaev Kamaru Usman	Middleweight	M-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Kamaru Usman Khamzat Chimaev	[27, 52, 27, 53]	[6, 10, 3, 9]	[3, 4, 8, 8]	[33, 63, 22, 53]	...	33	22	1	0	2	16	1	12	00:00:04	00:07:16
1	Trevor Peek Mohammad Yahya	Lightweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Mohammad Yahya Trevor Peek	[28, 70, 32, 61]	[14, 21, 16, 23]	[1, 1, 13, 13]	[41, 90, 50, 83]	...	41	50	2	6	0	5	1	7	00:00:24	00:05:10
2	Sedriques Dumas Abu Azaitar	Middleweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Abu Azaitar Sedriques Dumas	[21, 78, 11, 45]	[13, 18, 5, 11]	[7, 10, 18, 23]	[31, 90, 31, 76]	...	31	31	3	0	7	3	4	2	00:00:32	00:04:10
3	Muhammad Naimov Nathaniel Wood	Featherweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Nathaniel Wood Muhammad Naimov	[37, 57, 31, 60]	[6, 7, 10, 13]	[7, 8, 7, 8]	[36, 56, 24, 48]	...	36	24	5	11	9	13	7	9	00:02:44	00:06:42
4	Viktoriia Dudakova Jinh Yu Frey	Women's Strawweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Viktoriia Dudakova Jinh Yu Frey	[47, 115, 31, 80]	[26, 30, 8, 13]	[1, 1, 8, 8]	[60, 131, 37, 89]	...	60	37	14	3	0	7	3	3	00:00:18	00:04:40

5 rows × 40 columns

decision_data['Fighter_1_Ctrl'] = pd.to_timedelta(decision_data['Fighter_1_Ctrl'])
decision_data['Fighter_2_Ctrl'] = pd.to_timedelta(decision_data['Fighter_2_Ctrl'])

# Convert timedelta to total seconds and then to integers
decision_data['Fighter_1_Ctrl'] = decision_data['Fighter_1_Ctrl'].dt.total_seconds().astype(int)
decision_data['Fighter_2_Ctrl'] = decision_data['Fighter_2_Ctrl'].dt.total_seconds().astype(int)

decision_data.head()

	Fighter	Weight class	Method	Round	Event	Fighter_1	Head	Body	Leg	Distance	...	Fighter_1_Distance	Fighter_2_Distance	Fighter_1_Clinch	Fighter_2_Clinch	Fighter_1_Ground	Fighter_2_Ground	Fighter_1_Td_Att	Fighter_2_Td_Att	Fighter_1_Ctrl	Fighter_2_Ctrl
0	Khamzat Chimaev Kamaru Usman	Middleweight	M-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Kamaru Usman Khamzat Chimaev	[27, 52, 27, 53]	[6, 10, 3, 9]	[3, 4, 8, 8]	[33, 63, 22, 53]	...	33	22	1	0	2	16	1	12	4	436
1	Trevor Peek Mohammad Yahya	Lightweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Mohammad Yahya Trevor Peek	[28, 70, 32, 61]	[14, 21, 16, 23]	[1, 1, 13, 13]	[41, 90, 50, 83]	...	41	50	2	6	0	5	1	7	24	310
2	Sedriques Dumas Abu Azaitar	Middleweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Abu Azaitar Sedriques Dumas	[21, 78, 11, 45]	[13, 18, 5, 11]	[7, 10, 18, 23]	[31, 90, 31, 76]	...	31	31	3	0	7	3	4	2	32	250
3	Muhammad Naimov Nathaniel Wood	Featherweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Nathaniel Wood Muhammad Naimov	[37, 57, 31, 60]	[6, 7, 10, 13]	[7, 8, 7, 8]	[36, 56, 24, 48]	...	36	24	5	11	9	13	7	9	164	402
4	Viktoriia Dudakova Jinh Yu Frey	Women's Strawweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Viktoriia Dudakova Jinh Yu Frey	[47, 115, 31, 80]	[26, 30, 8, 13]	[1, 1, 8, 8]	[60, 131, 37, 89]	...	60	37	14	3	0	7	3	3	18	280

5 rows × 40 columns

# Converting columns to numeric

columns_to_convert = ["Fighter_1_Head", "Fighter_2_Head",
                     "Fighter_1_Body", "Fighter_2_Leg",
                     "Fighter_1_Distance", "Fighter_2_Distance",
                     "Fighter_1_Clinch", "Fighter_2_Clinch",
                     "Fighter_1_Ground", "Fighter_2_Ground"]
                     

decision_data[columns_to_convert] = decision_data[columns_to_convert].apply(pd.to_numeric)

decision_data.dtypes

Fighter               object
Weight class          object
Method                object
Round                  int64
Event                 object
Fighter_1             object
Head                  object
Body                  object
Leg                   object
Distance              object
Clinch                object
Ground                object
Sig. str.             object
Td_y                  object
Rev.                  object
Ctrl                  object
Fighter_1_Kd           int64
Fighter_2_Kd           int64
Fighter_1_Str          int64
Fighter_2_Str          int64
Fighter_1_Td           int64
Fighter_2_Td           int64
Fighter_1_Sub          int64
Fighter_2_Sub          int64
Fighter_1_Head         int64
Fighter_2_Head         int64
Fighter_1_Body         int64
Fighter_2_Body         int64
Fighter_1_Leg          int64
Fighter_2_Leg          int64
Fighter_1_Distance     int64
Fighter_2_Distance     int64
Fighter_1_Clinch       int64
Fighter_2_Clinch       int64
Fighter_1_Ground       int64
Fighter_2_Ground       int64
Fighter_1_Td_Att       int64
Fighter_2_Td_Att       int64
Fighter_1_Ctrl         int64
Fighter_2_Ctrl         int64
dtype: object

Now, I extract differentials for Fighter_1, which will be used to build the model.

decision_data["Diff_Kd"] = decision_data["Fighter_1_Kd"] - decision_data["Fighter_2_Kd"]
decision_data["Diff_Str"] = decision_data["Fighter_1_Str"] - decision_data["Fighter_2_Str"]
decision_data["Diff_Td"] = decision_data["Fighter_1_Td"] - decision_data["Fighter_2_Td"]
decision_data["Diff_Sub"] = decision_data["Fighter_1_Sub"] - decision_data["Fighter_2_Sub"]
decision_data["Diff_Head"] = decision_data["Fighter_1_Head"] - decision_data["Fighter_2_Head"]
decision_data["Diff_Body"] = decision_data["Fighter_1_Body"] - decision_data["Fighter_2_Body"]
decision_data["Diff_Leg"]= decision_data["Fighter_1_Leg"] - decision_data["Fighter_2_Leg"]
decision_data["Diff_Distance"] = decision_data["Fighter_1_Distance"] - decision_data["Fighter_2_Distance"]
decision_data["Diff_Clinch"] = decision_data["Fighter_1_Clinch"] - decision_data["Fighter_2_Clinch"]
decision_data["Diff_Ground"] = decision_data["Fighter_1_Ground"] - decision_data["Fighter_2_Ground"]
decision_data["Diff_Ctrl"] = decision_data["Fighter_1_Ctrl"] - decision_data["Fighter_2_Ctrl"]

Now, I create a column indicating if "Fighter_1" won the fight.

decision_data['Fighter_1_W'] = decision_data.apply(lambda row: 1 if row['Fighter_1'].replace(" ", "") == row['Fighter'].replace(" ", "") else 0, axis=1)
decision_data.head()

	Fighter	Weight class	Method	Round	Event	Fighter_1	Head	Body	Leg	Distance	...	Diff_Td	Diff_Sub	Diff_Head	Diff_Body	Diff_Leg	Diff_Distance	Diff_Clinch	Diff_Ground	Diff_Ctrl	Fighter_1_W
0	Khamzat Chimaev Kamaru Usman	Middleweight	M-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Kamaru Usman Khamzat Chimaev	[27, 52, 27, 53]	[6, 10, 3, 9]	[3, 4, 8, 8]	[33, 63, 22, 53]	...	-4	-1	0	3	-5	11	1	-14	-432	0
1	Trevor Peek Mohammad Yahya	Lightweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Mohammad Yahya Trevor Peek	[28, 70, 32, 61]	[14, 21, 16, 23]	[1, 1, 13, 13]	[41, 90, 50, 83]	...	-3	1	-4	-2	-12	-9	-4	-5	-286	0
2	Sedriques Dumas Abu Azaitar	Middleweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Abu Azaitar Sedriques Dumas	[21, 78, 11, 45]	[13, 18, 5, 11]	[7, 10, 18, 23]	[31, 90, 31, 76]	...	1	0	10	8	-11	0	3	4	-218	0
3	Muhammad Naimov Nathaniel Wood	Featherweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Nathaniel Wood Muhammad Naimov	[37, 57, 31, 60]	[6, 7, 10, 13]	[7, 8, 7, 8]	[36, 56, 24, 48]	...	-2	0	6	-4	0	12	-6	-4	-238	0
4	Viktoriia Dudakova Jinh Yu Frey	Women's Strawweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Viktoriia Dudakova Jinh Yu Frey	[47, 115, 31, 80]	[26, 30, 8, 13]	[1, 1, 8, 8]	[60, 131, 37, 89]	...	1	0	16	18	-7	23	11	-7	-262	1

5 rows × 52 columns

decision_data["First_Fighter"] = decision_data["Fighter_1"].apply(lambda x: " ".join(x.split()[:2]))

fighter_1 = decision_data[["Method", "Weight class", "Diff_Kd", "Diff_Str",
                           "Diff_Td", "Diff_Sub", "Diff_Head",
                           "Diff_Body", "Diff_Leg", "Diff_Distance",
                           "Diff_Clinch", "Diff_Ground", "Diff_Ctrl",
                           'Fighter_1_W']]

fighter_1.head()

	Method	Weight class	Diff_Kd	Diff_Str	Diff_Td	Diff_Sub	Diff_Head	Diff_Body	Diff_Leg	Diff_Distance	Diff_Clinch	Diff_Ground	Diff_Ctrl	Fighter_1_W
0	M-DEC	Middleweight	0	-2	-4	-1	0	3	-5	11	1	-14	-432	0
1	U-DEC	Lightweight	0	-18	-3	1	-4	-2	-12	-9	-4	-5	-286	0
2	U-DEC	Middleweight	0	7	1	0	10	8	-11	0	3	4	-218	0
3	U-DEC	Featherweight	0	2	-2	0	6	-4	0	12	-6	-4	-238	0
4	U-DEC	Women's Strawweight	-1	-27	1	0	16	18	-7	23	11	-7	-262	1

fighter_1.to_csv("model_data_1.csv")

fighter_1.dtypes

Method           object
Weight class     object
Diff_Kd           int64
Diff_Str          int64
Diff_Td           int64
Diff_Sub          int64
Diff_Head         int64
Diff_Body         int64
Diff_Leg          int64
Diff_Distance     int64
Diff_Clinch       int64
Diff_Ground       int64
Diff_Ctrl         int64
Fighter_1_W       int64
dtype: object

Before cleaning the data, it is time to extract the data that will be used for analysis. The fights that will be examined are fights that ended in a decision. Let's check all of the different ways fights have ended.

Decision fights are labelled as "U-DEC", "S-DEC", or "M-DEC". These fights are now extracted and placed in a new dataframe.

# Creating a new dataframe with fights that ended in a decision

decision_data = data_1[data_1['Method'].str.contains('DEC', case=False)].reset_index(drop=True)
decision_data.head()

	Fighter	Kd	Str	Td	Sub	Weight class	Method	Round	Event	key
0	Khamzat Chimaev Kamaru Usman	[0, 0]	[38, 36]	[4, 0]	[1, 0]	Middleweight	M-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Chimaev Kamaru Khamzat Usman
1	Trevor Peek Mohammad Yahya	[0, 0]	[61, 43]	[4, 1]	[0, 1]	Lightweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Mohammad Peek Trevor Yahya
2	Sedriques Dumas Abu Azaitar	[0, 0]	[34, 41]	[1, 2]	[0, 0]	Middleweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Abu Azaitar Dumas Sedriques
3	Muhammad Naimov Nathaniel Wood	[0, 0]	[48, 50]	[4, 2]	[0, 0]	Featherweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Muhammad Naimov Nathaniel Wood
4	Viktoriia Dudakova Jinh Yu Frey	[1, 0]	[74, 47]	[0, 1]	[0, 0]	Women's Strawweight	U-DEC	3	UFC 294: Makhachev vs. Volkanovski 2	Dudakova Frey Jinh Viktoriia Yu
...	...	...	...	...	...	...	...	...	...	...
3383	Dan Severn Ken Shamrock	[0, 0]	[16, 18]	[0, 0]	[0, 0]	Open Weight	S-DEC	3	UFC 9: Motor City Madness	Dan Ken Severn Shamrock
3384	Paul Varelans Joe Moreira	[0, 0]	[9, 12]	[0, 0]	[0, 0]	Open Weight	U-DEC	1	UFC 8: David vs Goliath	Joe Moreira Paul Varelans
3385	Dan Severn Oleg Taktarov	[0, 0]	[24, 8]	[0, 0]	[0, 1]	Open Weight	U-DEC	2	UFC - Ultimate Ultimate '95	Dan Oleg Severn Taktarov
3386	Oleg Taktarov Marco Ruas	[0, 0]	[6, 14]	[0, 0]	[1, 0]	Open Weight	U-DEC	1	UFC - Ultimate Ultimate '95	Marco Oleg Ruas Taktarov
3387	Dan Severn David Abbott	[0, 0]	[57, 3]	[1, 0]	[0, 0]	Open Weight	U-DEC	1	UFC - Ultimate Ultimate '95	Abbott Dan David Severn

3388 rows × 10 columns

decision_data.dtypes

Fighter         object
Kd              object
Str             object
Td              object
Sub             object
Weight class    object
Method          object
Round            int64
Event           object
key             object
dtype: object

EDA

For the EDA, I will modify the original dataset.

data_1 = pd.read_csv("ufc_initial_file.csv")
data_1.head()

	Unnamed: 0	W/L	Fighter	Kd	Str	Td	Sub	Weight class	Method	Round	Time	Event
0	0	win	Islam Makhachev Alexander Volkanovski	1 0	24 4	0 0	0 0	Lightweight	KO/TKO Kick	1	3:06	UFC 294: Makhachev vs. Volkanovski 2
1	1	win	Khamzat Chimaev Kamaru Usman	0 0	38 36	4 0	1 0	Middleweight	M-DEC	3	5:00	UFC 294: Makhachev vs. Volkanovski 2
2	2	nc nc	Magomed Ankalaev Johnny Walker	0 0	13 14	1 0	0 0	Light Heavyweight	CNC	1	3:13	UFC 294: Makhachev vs. Volkanovski 2
3	3	win	Ikram Aliskerov Warlley Alves	1 0	26 12	0 0	0 0	Middleweight	KO/TKO Punches	1	2:07	UFC 294: Makhachev vs. Volkanovski 2
4	4	win	Said Nurmagomedov Muin Gafurov	0 0	5 1	0 0	1 0	Bantamweight	SUB Guillotine Choke	1	1:13	UFC 294: Makhachev vs. Volkanovski 2

# Creating a new dataframe with fights that ended in a decision

decision_data_1 = data_1[data_1['Method'].str.contains('DEC', case=False)].reset_index(drop=True)
decision_data_1.head()

	Unnamed: 0	W/L	Fighter	Kd	Str	Td	Sub	Weight class	Method	Round	Time	Event
0	1	win	Khamzat Chimaev Kamaru Usman	0 0	38 36	4 0	1 0	Middleweight	M-DEC	3	5:00	UFC 294: Makhachev vs. Volkanovski 2
1	6	win	Trevor Peek Mohammad Yahya	0 0	61 43	4 1	0 1	Lightweight	U-DEC	3	5:00	UFC 294: Makhachev vs. Volkanovski 2
2	8	win	Sedriques Dumas Abu Azaitar	0 0	34 41	1 2	0 0	Middleweight	U-DEC	3	5:00	UFC 294: Makhachev vs. Volkanovski 2
3	10	win	Muhammad Naimov Nathaniel Wood	0 0	48 50	4 2	0 0	Featherweight	U-DEC	3	5:00	UFC 294: Makhachev vs. Volkanovski 2
4	11	win	Viktoriia Dudakova Jinh Yu Frey	1 0	74 47	0 1	0 0	Women's Strawweight	U-DEC	3	5:00	UFC 294: Makhachev vs. Volkanovski 2

In order to continue, I have to remove the non-numeric values.

decision_data_1.replace("--", 0, inplace = True)

decision_data_1[['Winner_Kd', 'Loser_Kd']] = decision_data_1['Kd'].str.split(' ', n=1, expand=True)
decision_data_1[['Winner_Str', 'Loser_Str']]= decision_data_1['Str'].str.split(' ', n=1, expand=True)
decision_data_1[['Winner_Td', 'Loser_Td']] = decision_data_1['Td'].str.split(' ', n=1, expand=True)
decision_data_1[['Winner_Sub', 'Loser_Sub']]= decision_data_1['Sub'].str.split(' ', n=1, expand=True)
decision_data_1.head()

	Unnamed: 0	W/L	Fighter	Kd	Str	Td	Sub	Weight class	Method	Round	Time	Event	Winner_Kd	Loser_Kd	Winner_Str	Loser_Str	Winner_Td	Loser_Td	Winner_Sub	Loser_Sub
0	1	win	Khamzat Chimaev Kamaru Usman	0 0	38 36	4 0	1 0	Middleweight	M-DEC	3	5:00	UFC 294: Makhachev vs. Volkanovski 2	0	0	38	36	4	0	1	0
1	6	win	Trevor Peek Mohammad Yahya	0 0	61 43	4 1	0 1	Lightweight	U-DEC	3	5:00	UFC 294: Makhachev vs. Volkanovski 2	0	0	61	43	4	1	0	1
2	8	win	Sedriques Dumas Abu Azaitar	0 0	34 41	1 2	0 0	Middleweight	U-DEC	3	5:00	UFC 294: Makhachev vs. Volkanovski 2	0	0	34	41	1	2	0	0
3	10	win	Muhammad Naimov Nathaniel Wood	0 0	48 50	4 2	0 0	Featherweight	U-DEC	3	5:00	UFC 294: Makhachev vs. Volkanovski 2	0	0	48	50	4	2	0	0
4	11	win	Viktoriia Dudakova Jinh Yu Frey	1 0	74 47	0 1	0 0	Women's Strawweight	U-DEC	3	5:00	UFC 294: Makhachev vs. Volkanovski 2	1	0	74	47	0	1	0	0

decision_data_1.replace("--", 0, inplace=True)
decision_data_1.replace(" --", 0, inplace=True)
decision_data_1[["Winner_Str", "Loser_Str", "Winner_Kd", "Loser_Kd", "Winner_Td", "Loser_Td", "Winner_Sub", "Loser_Sub"]] = decision_data_1[["Winner_Str", "Loser_Str", "Winner_Kd", "Loser_Kd", "Winner_Td", "Loser_Td", "Winner_Sub", "Loser_Sub"]].apply(pd.to_numeric)
decision_data_1.head()

	Unnamed: 0	W/L	Fighter	Kd	Str	Td	Sub	Weight class	Method	Round	Time	Event	Winner_Kd	Loser_Kd	Winner_Str	Loser_Str	Winner_Td	Loser_Td	Winner_Sub	Loser_Sub
0	1	win	Khamzat Chimaev Kamaru Usman	0 0	38 36	4 0	1 0	Middleweight	M-DEC	3	5:00	UFC 294: Makhachev vs. Volkanovski 2	0	0	38	36	4	0	1	0
1	6	win	Trevor Peek Mohammad Yahya	0 0	61 43	4 1	0 1	Lightweight	U-DEC	3	5:00	UFC 294: Makhachev vs. Volkanovski 2	0	0	61	43	4	1	0	1
2	8	win	Sedriques Dumas Abu Azaitar	0 0	34 41	1 2	0 0	Middleweight	U-DEC	3	5:00	UFC 294: Makhachev vs. Volkanovski 2	0	0	34	41	1	2	0	0
3	10	win	Muhammad Naimov Nathaniel Wood	0 0	48 50	4 2	0 0	Featherweight	U-DEC	3	5:00	UFC 294: Makhachev vs. Volkanovski 2	0	0	48	50	4	2	0	0
4	11	win	Viktoriia Dudakova Jinh Yu Frey	1 0	74 47	0 1	0 0	Women's Strawweight	U-DEC	3	5:00	UFC 294: Makhachev vs. Volkanovski 2	1	0	74	47	0	1	0	0

decision_data_1["Diff_Kd"] = decision_data_1["Winner_Kd"] - decision_data_1["Loser_Kd"]
decision_data_1["Diff_Str"] = decision_data_1["Winner_Str"] - decision_data_1["Loser_Str"]
decision_data_1["Diff_Td"] = decision_data_1["Winner_Td"] - decision_data_1["Loser_Td"]
decision_data_1["Diff_Sub"] = decision_data_1["Winner_Sub"] - decision_data_1["Loser_Sub"]
decision_data_1.head()

	Unnamed: 0	W/L	Fighter	Kd	Str	Td	Sub	Weight class	Method	Round	...	Winner_Str	Loser_Str	Winner_Td	Loser_Td	Winner_Sub	Loser_Sub	Diff_Kd	Diff_Str	Diff_Td	Diff_Sub
0	1	win	Khamzat Chimaev Kamaru Usman	0 0	38 36	4 0	1 0	Middleweight	M-DEC	3	...	38	36	4	0	1	0	0	2	4	1
1	6	win	Trevor Peek Mohammad Yahya	0 0	61 43	4 1	0 1	Lightweight	U-DEC	3	...	61	43	4	1	0	1	0	18	3	-1
2	8	win	Sedriques Dumas Abu Azaitar	0 0	34 41	1 2	0 0	Middleweight	U-DEC	3	...	34	41	1	2	0	0	0	-7	-1	0
3	10	win	Muhammad Naimov Nathaniel Wood	0 0	48 50	4 2	0 0	Featherweight	U-DEC	3	...	48	50	4	2	0	0	0	-2	2	0
4	11	win	Viktoriia Dudakova Jinh Yu Frey	1 0	74 47	0 1	0 0	Women's Strawweight	U-DEC	3	...	74	47	0	1	0	0	1	27	-1	0

5 rows × 24 columns

decision_data_1 = decision_data_1[(decision_data_1["Round"] == 3) | (decision_data_1["Round"] == 5)]

A cursory analysis of fights that ended in a decision can offer insight into what predicts the winner of such a fight. A good place to start is with Unanimous Decisions. Perhaps by looking at fights where every judge agreed on choosing the winner, it will become obvious as to what a fighter must do to win a fight. The chart below illustrates the distribution in the winning fighter's significant strike differential. As seen in the chart, the distribution is right-skewed with a mean of 22.65 and 26.29. Hence, fighters who won convincingly usually landed more significant strikes than their opponent; however, the chart does show that a considerable number of fighters who won unanimous decisions lost the striking battle.

# Visualize Distribution of Winner Strike Differential in Unanimous Decision Fights

U_DEC = decision_data_1[decision_data_1["Method"] == "U-DEC"]
plt.hist(U_DEC["Diff_Str"], bins = 40)
mean_value = U_DEC["Diff_Str"].mean()
std_value = U_DEC["Diff_Str"].std()
greater_than_0 = (len(U_DEC[U_DEC["Diff_Str"] > 0])/len(U_DEC))*100
less_than_0 = (len(U_DEC[U_DEC["Diff_Str"] < 0])/len(U_DEC))*100
plt.text(0.85, 0.9, f'Mean: {mean_value:.2f}', fontsize=12, transform=plt.gca().transAxes, va='center', ha='center')
plt.text(0.85, 0.8, f'Std: {std_value:.2f}', fontsize=12, transform=plt.gca().transAxes, va='center', ha='center')
plt.text(0.75, 0.7, f'% Greater Than 0: {greater_than_0:.2f}', fontsize=12, transform=plt.gca().transAxes, va='center', ha='center')
plt.text(0.75, 0.6, f'% Less Than 0: {less_than_0:.2f}', fontsize=12, transform=plt.gca().transAxes, va='center', ha='center')
plt.axvline(x=0, color='red', linestyle='--', linewidth=2, label='Zero')
plt.xlabel('Significant Strike Differential')
plt.ylabel('Frequency')
plt.title('Unanimous Decision Winner Sig. Strike Differential (Per 3 Rounds)')
plt.show()

The chart below shows the distribution of the winning fighter's takedown differential. The distribution is slightly right-skewed with a mean of 1.55 and standard deviation of 2.98. The chart suggests that winners of a unanimous decision are more likely to win the takedown battle, but, as with significant strikes, quite a few winning fighters lost the takedown battle.

# Visualize Distribution of Winner Takedown Differential in Unanimous Decision Fights
plt.figure(figsize=(9,4))
plt.hist(U_DEC["Diff_Td"], bins = 30)
mean_value = U_DEC["Diff_Td"].mean()
std_value =  U_DEC["Diff_Td"].std()
greater_than_0 = (len(U_DEC[U_DEC["Diff_Td"] > 0])/len(U_DEC))*100
less_than_0 = (len(U_DEC[U_DEC["Diff_Td"] < 0])/len(U_DEC))*100
plt.text(0.85, 0.9, f'Mean: {mean_value:.2f}', fontsize=12, transform=plt.gca().transAxes, va='center', ha='center')
plt.text(0.85, 0.8, f'Std: {std_value:.2f}', fontsize=12, transform=plt.gca().transAxes, va='center', ha='center')
plt.text(0.85, 0.7, f'% Greater Than 0: {greater_than_0:.2f}', fontsize=12, transform=plt.gca().transAxes, va='center', ha='center')
plt.text(0.85, 0.6, f'% Less Than 0: {less_than_0:.2f}', fontsize=12, transform=plt.gca().transAxes, va='center', ha='center')
plt.axvline(x=0, color='red', linestyle='--', linewidth=2, label='Zero')
plt.xlabel('Takedown Differential')
plt.ylabel('Frequency')
plt.title('Unanimous Decision Winner Takedown Differential (Per 3 Rounds)')
plt.show()

Split Decision Analysis

Split Decisions occur when one of the three judges scores the declared winner of the fight as the loser. As there is no consensus winner among the judges, these fights are inherently controversial. Thus, it is to be expected that the mean strike differential is to be smaller than those of a Unanimous Decision. The chart below bears this out. The distribution has little-to-no skew with a mean of 4.46 and 20.38 standard deviation. Additionally, many winners of split decisions actually lost the striking battle.

# Visualize Distribution of Winner Strike Differential in Split Decision Fights

S_DEC = decision_data_1[decision_data_1["Method"] == "S-DEC"]
plt.hist(S_DEC["Diff_Str"], bins = 30)
mean_value = S_DEC["Diff_Str"].mean()
std_value = S_DEC["Diff_Str"].std()
greater_than_0 = (len(S_DEC[S_DEC["Diff_Str"] > 0])/len(S_DEC))*100
less_than_0 = (len(S_DEC[S_DEC["Diff_Str"] < 0])/len(S_DEC))*100
plt.text(0.85, 0.9, f'Mean: {mean_value:.2f}', fontsize=12, transform=plt.gca().transAxes, va='center', ha='center')
plt.text(0.85, 0.8, f'Std: {std_value:.2f}', fontsize=12, transform=plt.gca().transAxes, va='center', ha='center')
plt.text(0.75, 0.7, f'% Greater Than 0: {greater_than_0:.2f}', fontsize=12, transform=plt.gca().transAxes, va='center', ha='center')
plt.text(0.75, 0.6, f'% Less Than 0: {less_than_0:.2f}', fontsize=12, transform=plt.gca().transAxes, va='center', ha='center')
plt.axvline(x=0, color='red', linestyle='--', linewidth=2, label='Zero')
plt.xlabel("Significant Strike Differential")
plt.ylabel('Frequency')
plt.title('Split Decision Winner Sig. Strike Differential (Per 3 Rounds)')
plt.show()

In a similar fashion to the chart above, the distribution of the takedown differential has almost no skew. The mean is 0.23, and the standard deviation is 2.63. These two charts suggest that perhaps strike differential and takedown differential are not excellent predictors for the winner of a split decision. Perhaps more advanced striking and takedown data can assist in painting a higher-resolution picture of these fights.

# Visualize Distribution of Winner Takedown Differential in Split Decision Fights
plt.figure(figsize=(8, 5))
plt.hist(S_DEC["Diff_Td"], bins = 15)
mean_value = S_DEC["Diff_Td"].mean()
std_value = S_DEC["Diff_Td"].std()
greater_than_0 = (len(S_DEC[S_DEC["Diff_Td"] > 0])/len(S_DEC))*100
less_than_0 = (len(S_DEC[S_DEC["Diff_Td"] < 0])/len(S_DEC))*100
plt.text(0.85, 0.9, f'Mean: {mean_value:.2f}', fontsize=12, transform=plt.gca().transAxes, va='center', ha='center')
plt.text(0.85, 0.8, f'Std: {std_value:.2f}', fontsize=12, transform=plt.gca().transAxes, va='center', ha='center')
plt.text(0.81, 0.7, f'% Greater than 0: {greater_than_0:.2f}', fontsize=12, transform=plt.gca().transAxes, va='center', ha='center')
plt.text(0.8, 0.6, f'% Less than 0: {less_than_0:.2f}', fontsize=12, transform=plt.gca().transAxes, va='center', ha='center')
plt.axvline(x=0, color='red', linestyle='--', linewidth=2, label='Zero')
plt.xlabel('Takedown Differential')
plt.ylabel('Frequency')
plt.title('Split Decision Winner Takedown Differential (Per 3 Rounds)')
plt.show()

What can be gleaned from this cursory analysis? In split decisions, it appears that Takedown Differential and Significant Strikes Differential will not be significant predictors in the model. It appears that some more "advanced" metrics will be necessary to include when building out the model.

Model Selection

The goal of the model is to accurately predict fights and offer insight about the influence of features. As such, a logistic regression model has been chosen.

Model for All Types of Fights

The goal of the first model is to predict the result of a fight without knowing anything about the type of decision. This model will provide a broad idea of what features are important when predicting any fight.

from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
import statsmodels.api as sm
from statsmodels.tools import add_constant

# Import the Data

data = pd.read_csv("model_data_1.csv")
data.head()

	Unnamed: 0	Method	Weight class	Diff_Kd	Diff_Str	Diff_Td	Diff_Sub	Diff_Head	Diff_Body	Diff_Leg	Diff_Distance	Diff_Clinch	Diff_Ground	Diff_Ctrl	Fighter_1_W
0	0	M-DEC	Middleweight	0	-2	-4	-1	0	3	-5	11	1	-14	-432	0
1	1	U-DEC	Lightweight	0	-18	-3	1	-4	-2	-12	-9	-4	-5	-286	0
2	2	U-DEC	Middleweight	0	7	1	0	10	8	-11	0	3	4	-218	0
3	3	U-DEC	Featherweight	0	2	-2	0	6	-4	0	12	-6	-4	-238	0
4	4	U-DEC	Women's Strawweight	-1	-27	1	0	16	18	-7	23	11	-7	-262	1

# Removing unnecessary columns

data = data.drop(data.columns[0], axis = 1)
data.head()

	Method	Weight class	Diff_Kd	Diff_Str	Diff_Td	Diff_Sub	Diff_Head	Diff_Body	Diff_Leg	Diff_Distance	Diff_Clinch	Diff_Ground	Diff_Ctrl	Fighter_1_W
0	M-DEC	Middleweight	0	-2	-4	-1	0	3	-5	11	1	-14	-432	0
1	U-DEC	Lightweight	0	-18	-3	1	-4	-2	-12	-9	-4	-5	-286	0
2	U-DEC	Middleweight	0	7	1	0	10	8	-11	0	3	4	-218	0
3	U-DEC	Featherweight	0	2	-2	0	6	-4	0	12	-6	-4	-238	0
4	U-DEC	Women's Strawweight	-1	-27	1	0	16	18	-7	23	11	-7	-262	1

Checking for Multicollinearity

Before building the model, it is important to ensure that there is not multicollinearity within the data. First, a correlation matrix will be analyzed.

# Creating the Correlation Matrix

correlation_matrix = data[["Diff_Kd", "Diff_Str", "Diff_Td", "Diff_Sub", "Diff_Head",
                          "Diff_Body", "Diff_Leg", "Diff_Distance", "Diff_Clinch",
                          "Diff_Ground", "Diff_Ctrl"]].corr()

plt.figure(figsize=(8, 6))
sns.heatmap(correlation_matrix, annot=True, cmap='coolwarm', fmt='.2f', linewidths=.5)
plt.title('Correlation Matrix')
plt.show()

Head Strikes Differential and Distance Strikes Differential are highly correlated, so Distance Strikes Differential is removed from the data used for modeling. Now, VIF scores are checked.

# Finding VIF Scores

from statsmodels.stats.outliers_influence import variance_inflation_factor
data_with_const = sm.add_constant(data[["Diff_Kd", "Diff_Str", "Diff_Td", "Diff_Sub", "Diff_Head",
                          "Diff_Body", "Diff_Leg", "Diff_Clinch",
                          "Diff_Ground", "Diff_Ctrl"]])

vif_data = pd.DataFrame()
vif_data["Variable"] = data_with_const.columns
vif_data["VIF"] = [variance_inflation_factor(data_with_const.values, i) for i in range(data_with_const.shape[1])]

print(vif_data)

       Variable       VIF
0         const  1.755790
1       Diff_Kd  1.008553
2      Diff_Str  1.070263
3       Diff_Td  1.064902
4      Diff_Sub  1.005223
5     Diff_Head  1.646237
6     Diff_Body  1.354418
7      Diff_Leg  1.084286
8   Diff_Clinch  1.476568
9   Diff_Ground  2.235665
10    Diff_Ctrl  1.737613

The VIF scores are all low, and so multicollinearity is not a concern. Now, the model can be built.

# Creating the training and testing set

import sklearn
from sklearn.model_selection import train_test_split

train, test = train_test_split(data,
                               test_size=0.3,
                               stratify=data['Method'],
                               random_state = 42)

# Adjusting the training and testing sets

X_train = train[["Diff_Kd", "Diff_Str", "Diff_Td", "Diff_Sub", "Diff_Head", "Diff_Body", "Diff_Leg", 
            "Diff_Clinch", "Diff_Ground", "Diff_Ctrl"]]
y_train = train["Fighter_1_W"]
X_test = test[["Diff_Kd", "Diff_Str", "Diff_Td", "Diff_Sub", "Diff_Head", "Diff_Body", "Diff_Leg", 
            "Diff_Clinch", "Diff_Ground", "Diff_Ctrl"]]
y_test = test["Fighter_1_W"]

# Creating the Model

features_1 = ["Diff_Kd", "Diff_Str", "Diff_Td", "Diff_Sub", "Diff_Head", "Diff_Body", "Diff_Leg", 
            "Diff_Clinch", "Diff_Ground", "Diff_Ctrl"]
scaler = StandardScaler()
X_train_scaled = scaler.fit_transform(X_train)
X_train_with_intercept = add_constant(X_train_scaled)

logit_model = sm.Logit(y_train,X_train_with_intercept)
result_1 = logit_model.fit()
print(result_1.summary(xname=["const"] + features_1))

Optimization terminated successfully.
         Current function value: 0.346389
         Iterations 8
                           Logit Regression Results                           
==============================================================================
Dep. Variable:            Fighter_1_W   No. Observations:                 2312
Model:                          Logit   Df Residuals:                     2301
Method:                           MLE   Df Model:                           10
Date:                Tue, 12 Dec 2023   Pseudo R-squ.:                  0.4753
Time:                        13:02:14   Log-Likelihood:                -800.85
converged:                       True   LL-Null:                       -1526.4
Covariance Type:            nonrobust   LLR p-value:                8.855e-306
===============================================================================
                  coef    std err          z      P>|z|      [0.025      0.975]
-------------------------------------------------------------------------------
const           0.8660      0.083     10.458      0.000       0.704       1.028
Diff_Kd         0.1662      0.066      2.520      0.012       0.037       0.295
Diff_Str        0.3626      0.103      3.509      0.000       0.160       0.565
Diff_Td         0.2000      0.072      2.784      0.005       0.059       0.341
Diff_Sub       -0.0691      0.067     -1.027      0.304      -0.201       0.063
Diff_Head       1.9469      0.123     15.783      0.000       1.705       2.189
Diff_Body       0.6200      0.083      7.481      0.000       0.458       0.782
Diff_Leg        0.5947      0.070      8.481      0.000       0.457       0.732
Diff_Clinch    -0.0030      0.088     -0.034      0.973      -0.175       0.169
Diff_Ground     0.3204      0.146      2.194      0.028       0.034       0.607
Diff_Ctrl       1.3975      0.099     14.151      0.000       1.204       1.591
===============================================================================

As seen in the table above, 8 of the 10 predictors used are significant in the model, with Submission Attempts Differential and Clinch Strikes Differential being insignificant. Head Strikes Differential and Control Time Differential are the predictors with the highest coefficients in the model. Now, let's see the accuracy score of the model.

# Finding the Accuracy

X_test_scaled = scaler.transform(X_test)  # Use the same scaler as on the training set
X_test_with_intercept = add_constant(X_test_scaled)

# Predict on the test set
y_pred = result_1.predict(X_test_with_intercept)

# Convert predicted probabilities to binary predictions (0 or 1)
y_pred_binary = (y_pred > 0.5).astype(int)

# Calculate accuracy
accuracy_1 = (y_pred_binary == y_test).mean()
print(f"Accuracy: {accuracy_1:.2f}")

Accuracy: 0.86

Below, an amended model is made by making categorical variables from the insignificant predictors. If the fighter output in the insignificant features is greater than 95% (an arbitrarily chosen percentage) of the observations in that column, then that observation is changed to a 1. If the observation does not meet this criteria, it is changed to a 0.

# Creating New Training Dataframe

X_train_adj = X_train.copy()

third_quartile_value = data["Diff_Clinch"].quantile(0.95)
X_train_adj.loc[:, "Diff_Clinch"] = X_train_adj["Diff_Clinch"].apply(lambda x: 1 if x > third_quartile_value else 0)

third_quartile_value_1 = data["Diff_Sub"].quantile(0.95)
X_train_adj.loc[:, "Diff_Sub"] = X_train_adj["Diff_Sub"].apply(lambda x: 1 if x > third_quartile_value_1 else 0)

# Creating New Testing Dataframe

X_test_adj = X_test.copy()

third_quartile_value = data["Diff_Clinch"].quantile(0.95)
X_test_adj.loc[:, "Diff_Clinch"] = X_test_adj["Diff_Clinch"].apply(lambda x: 1 if x > third_quartile_value else 0)

third_quartile_value_1 = data["Diff_Sub"].quantile(0.95)
X_test_adj.loc[:, "Diff_Sub"] = X_test_adj["Diff_Sub"].apply(lambda x: 1 if x > third_quartile_value_1 else 0)

# Creating the Model

features = ["Diff_Kd", "Diff_Str", "Diff_Td", "Diff_Sub_Cat", "Diff_Head", "Diff_Body", "Diff_Leg", 
            "Diff_Clinch_Cat", "Diff_Ground", "Diff_Ctrl"]
scaler = StandardScaler()
X_train_adj_scaled = scaler.fit_transform(X_train_adj)
X_train_adj_scaled_with_intercept = add_constant(X_train_adj_scaled)
X_train_adj_scaled_with_intercept

logit_model_adj = sm.Logit(y_train, X_train_adj_scaled_with_intercept)
result_adj = logit_model_adj.fit()
print(result_adj.summary(xname=["const"] + features))

Optimization terminated successfully.
         Current function value: 0.346313
         Iterations 8
                           Logit Regression Results                           
==============================================================================
Dep. Variable:            Fighter_1_W   No. Observations:                 2312
Model:                          Logit   Df Residuals:                     2301
Method:                           MLE   Df Model:                           10
Date:                Tue, 12 Dec 2023   Pseudo R-squ.:                  0.4755
Time:                        13:02:20   Log-Likelihood:                -800.67
converged:                       True   LL-Null:                       -1526.4
Covariance Type:            nonrobust   LLR p-value:                7.435e-306
===================================================================================
                      coef    std err          z      P>|z|      [0.025      0.975]
-----------------------------------------------------------------------------------
const               0.8726      0.083     10.460      0.000       0.709       1.036
Diff_Kd             0.1654      0.066      2.503      0.012       0.036       0.295
Diff_Str            0.3731      0.103      3.607      0.000       0.170       0.576
Diff_Td             0.2065      0.072      2.872      0.004       0.066       0.347
Diff_Sub_Cat        0.0469      0.079      0.594      0.552      -0.108       0.201
Diff_Head           1.9339      0.119     16.317      0.000       1.702       2.166
Diff_Body           0.6062      0.076      7.952      0.000       0.457       0.756
Diff_Leg            0.5910      0.069      8.539      0.000       0.455       0.727
Diff_Clinch_Cat     0.1005      0.101      0.991      0.322      -0.098       0.299
Diff_Ground         0.3204      0.145      2.214      0.027       0.037       0.604
Diff_Ctrl           1.3933      0.099     14.138      0.000       1.200       1.586
===================================================================================

# Finding Accuracy

X_test_adj_scaled = scaler.transform(X_test_adj)  
X_test_adj_with_intercept = add_constant(X_test_adj_scaled)

y_pred = result_adj.predict(X_test_adj_with_intercept)

y_pred_binary = (y_pred > 0.5).astype(int)

# Calculate accuracy
accuracy_adj = (y_pred_binary == y_test).mean()
print(f"Accuracy: {accuracy_adj:.2f}")

Accuracy: 0.86

The same insignificatant predictors remain. The accuracy is still the same. I will now remove the remaining insignificant predictors.

# Creating the New Model

features_adj = ["Diff_Kd", "Diff_Str", "Diff_Td", "Diff_Head", "Diff_Body", "Diff_Leg", 
            "Diff_Ground", "Diff_Ctrl"]

X_train_adj.drop(["Diff_Sub", "Diff_Clinch"], axis=1, inplace=True)

scaler = StandardScaler()
X_train_adj_scaled = scaler.fit_transform(X_train_adj)
X_train_adj_scaled_with_intercept = add_constant(X_train_adj_scaled)
X_train_adj_scaled_with_intercept

logit_model_adj = sm.Logit(y_train, X_train_adj_scaled_with_intercept)
result_adj_1 = logit_model_adj.fit()
print(result_adj_1.summary(xname=["const"] + features_adj))

Optimization terminated successfully.
         Current function value: 0.346617
         Iterations 8
                           Logit Regression Results                           
==============================================================================
Dep. Variable:            Fighter_1_W   No. Observations:                 2312
Model:                          Logit   Df Residuals:                     2303
Method:                           MLE   Df Model:                            8
Date:                Tue, 12 Dec 2023   Pseudo R-squ.:                  0.4750
Time:                        13:02:24   Log-Likelihood:                -801.38
converged:                       True   LL-Null:                       -1526.4
Covariance Type:            nonrobust   LLR p-value:                8.244e-308
===============================================================================
                  coef    std err          z      P>|z|      [0.025      0.975]
-------------------------------------------------------------------------------
const           0.8637      0.083     10.448      0.000       0.702       1.026
Diff_Kd         0.1662      0.066      2.516      0.012       0.037       0.296
Diff_Str        0.3665      0.103      3.552      0.000       0.164       0.569
Diff_Td         0.2020      0.072      2.815      0.005       0.061       0.343
Diff_Head       1.9475      0.118     16.515      0.000       1.716       2.179
Diff_Body       0.6177      0.075      8.217      0.000       0.470       0.765
Diff_Leg        0.5936      0.069      8.586      0.000       0.458       0.729
Diff_Ground     0.3152      0.145      2.174      0.030       0.031       0.599
Diff_Ctrl       1.3941      0.099     14.153      0.000       1.201       1.587
===============================================================================

# Finding the Accuracy

X_test_adj.drop(["Diff_Sub", "Diff_Clinch"], axis=1, inplace=True)

X_test_adj_scaled = scaler.transform(X_test_adj)  
X_test_adj_with_intercept = add_constant(X_test_adj_scaled)

y_pred = result_adj_1.predict(X_test_adj_with_intercept)

y_pred_binary = (y_pred > 0.5).astype(int)

# Calculate accuracy
accuracy_adj_1 = (y_pred_binary == y_test).mean()
print(f"Accuracy: {accuracy_adj_1:.2f}")

Accuracy: 0.86

This model has the same accuracy as all of the other models, but it only contains significant predictors. As such, this model would be chosen.

As said, the model above is useful when only the fighter output is known. However, it would be useful to build a model on a subset of the data used for the model above. The subset will be based on the type of decision. In creating this model, more insights can be gained about the impact of features once more is known about the interpretation of the competition.

Unanimous Decision Model

First, a model will be created on fights that were scored a unanimous decision.

# Creating the training and testing sets

ud_train = train[train["Method"] == "U-DEC"]
ud_test = test[test["Method"] == "U-DEC"]

# Creating predictors and targets

ud_X_train = ud_train[["Diff_Kd", "Diff_Str", "Diff_Td", "Diff_Sub", "Diff_Head", "Diff_Body", "Diff_Leg", 
            "Diff_Clinch", "Diff_Ground", "Diff_Ctrl"]]
ud_y_train =ud_train["Fighter_1_W"]
ud_X_test = ud_test[["Diff_Kd", "Diff_Str", "Diff_Td", "Diff_Sub", "Diff_Head", "Diff_Body", "Diff_Leg", 
            "Diff_Clinch", "Diff_Ground", "Diff_Ctrl"]]
ud_y_test = ud_test["Fighter_1_W"]

# Building the model

features_2 = ["Diff_Kd", "Diff_Str", "Diff_Td", "Diff_Sub", "Diff_Head", "Diff_Body", "Diff_Leg", 
            "Diff_Clinch", "Diff_Ground", "Diff_Ctrl"]
scaler = StandardScaler()
ud_X_train_scaled = scaler.fit_transform(ud_X_train)
ud_X_train_with_intercept = add_constant(ud_X_train_scaled)

ud_logit_model = sm.Logit(ud_y_train, ud_X_train_with_intercept)
ud_result = ud_logit_model.fit()
print(ud_result.summary(xname=["const"] + features_2))

Optimization terminated successfully.
         Current function value: 0.233096
         Iterations 9
                           Logit Regression Results                           
==============================================================================
Dep. Variable:            Fighter_1_W   No. Observations:                 1778
Model:                          Logit   Df Residuals:                     1767
Method:                           MLE   Df Model:                           10
Date:                Tue, 12 Dec 2023   Pseudo R-squ.:                  0.6433
Time:                        13:02:33   Log-Likelihood:                -414.44
converged:                       True   LL-Null:                       -1161.7
Covariance Type:            nonrobust   LLR p-value:                     0.000
===============================================================================
                  coef    std err          z      P>|z|      [0.025      0.975]
-------------------------------------------------------------------------------
const           1.2235      0.146      8.392      0.000       0.938       1.509
Diff_Kd         0.1968      0.098      2.009      0.045       0.005       0.389
Diff_Str        0.5126      0.160      3.203      0.001       0.199       0.826
Diff_Td         0.2294      0.105      2.194      0.028       0.024       0.434
Diff_Sub       -0.0942      0.102     -0.920      0.357      -0.295       0.106
Diff_Head       2.9089      0.209     13.928      0.000       2.500       3.318
Diff_Body       0.8977      0.126      7.122      0.000       0.651       1.145
Diff_Leg        0.7792      0.100      7.799      0.000       0.583       0.975
Diff_Clinch    -0.2040      0.131     -1.554      0.120      -0.461       0.053
Diff_Ground     0.7814      0.255      3.058      0.002       0.281       1.282
Diff_Ctrl       1.7974      0.149     12.094      0.000       1.506       2.089
===============================================================================

In the model above, Submission Attempts Differential and Clinch Strikes Differential are the only insignificant predictors. Head Strikes Differential and Control Time Differential are still the most impactful features by a considerable margin.

# Finding the Accuracy

ud_X_test_scaled = scaler.transform(ud_X_test)  
ud_X_test_with_intercept = add_constant(ud_X_test_scaled)

ud_y_pred = ud_result.predict(ud_X_test_with_intercept)

ud_y_pred_binary = (ud_y_pred > 0.5).astype(int)

accuracy_2 = (ud_y_pred_binary == ud_y_test).mean()
print(f"Accuracy: {accuracy_2:.2f}")

Accuracy: 0.90

Below, an amended model is made by creating categorical variables in the same manner as the previos model.

# Creating New Training Dataframe

ud_X_train_adj = ud_X_train.copy()

third_quartile_value = data["Diff_Clinch"].quantile(0.95)
ud_X_train_adj.loc[:, "Diff_Clinch"] = ud_X_train_adj["Diff_Clinch"].apply(lambda x: 1 if x > third_quartile_value else 0)

third_quartile_value = data["Diff_Sub"].quantile(0.95)
ud_X_train_adj.loc[:, "Diff_Sub"] = ud_X_train_adj["Diff_Sub"].apply(lambda x: 1 if x > third_quartile_value else 0)

# Creating New Testing Dataframe

ud_X_test_adj = ud_X_test.copy()

third_quartile_value = data["Diff_Clinch"].quantile(0.95)
ud_X_test_adj.loc[:, "Diff_Clinch"] = ud_X_test_adj["Diff_Clinch"].apply(lambda x: 1 if x > third_quartile_value else 0)

third_quartile_value = data["Diff_Sub"].quantile(0.95)
ud_X_test_adj.loc[:, "Diff_Sub"] = ud_X_test_adj["Diff_Sub"].apply(lambda x: 1 if x > third_quartile_value else 0)

# Creating New Model

features_3 = ["Diff_Kd", "Diff_Str", "Diff_Td", "Diff_Sub_Cat", "Diff_Head", "Diff_Body", "Diff_Leg", 
            "Diff_Clinch_Cat", "Diff_Ground", "Diff_Ctrl"]
scaler = StandardScaler()
ud_X_train_adj_scaled = scaler.fit_transform(ud_X_train_adj)
ud_X_train_adj_with_intercept = add_constant(ud_X_train_adj_scaled)

ud_logit_model_adj = sm.Logit(ud_y_train, ud_X_train_adj_with_intercept)
ud_result_adj = ud_logit_model_adj.fit()
print(ud_result_adj.summary(xname=["const"] + features_3))

Optimization terminated successfully.
         Current function value: 0.233651
         Iterations 8
                           Logit Regression Results                           
==============================================================================
Dep. Variable:            Fighter_1_W   No. Observations:                 1778
Model:                          Logit   Df Residuals:                     1767
Method:                           MLE   Df Model:                           10
Date:                Tue, 12 Dec 2023   Pseudo R-squ.:                  0.6424
Time:                        13:02:40   Log-Likelihood:                -415.43
converged:                       True   LL-Null:                       -1161.7
Covariance Type:            nonrobust   LLR p-value:                     0.000
===================================================================================
                      coef    std err          z      P>|z|      [0.025      0.975]
-----------------------------------------------------------------------------------
const               1.2198      0.146      8.341      0.000       0.933       1.506
Diff_Kd             0.1791      0.097      1.838      0.066      -0.012       0.370
Diff_Str            0.5444      0.160      3.396      0.001       0.230       0.859
Diff_Td             0.2424      0.105      2.303      0.021       0.036       0.449
Diff_Sub_Cat        0.0814      0.139      0.585      0.559      -0.191       0.354
Diff_Head           2.8149      0.199     14.141      0.000       2.425       3.205
Diff_Body           0.7904      0.112      7.060      0.000       0.571       1.010
Diff_Leg            0.7477      0.098      7.667      0.000       0.557       0.939
Diff_Clinch_Cat     0.1495      0.157      0.953      0.341      -0.158       0.457
Diff_Ground         0.7659      0.252      3.045      0.002       0.273       1.259
Diff_Ctrl           1.8119      0.149     12.121      0.000       1.519       2.105
===================================================================================

None of the insignificant predictors become significant.

# Finding Accuracy

ud_X_test_scaled_adj = scaler.transform(ud_X_test_adj)  
ud_X_test_with_intercept_adj = add_constant(ud_X_test_scaled_adj)

ud_y_pred = ud_result_adj.predict(ud_X_test_with_intercept_adj)

ud_y_pred_binary = (ud_y_pred > 0.5).astype(int)

accuracy_3 = (ud_y_pred_binary == ud_y_test).mean()
print(f"Accuracy: {accuracy_3:.2f}")

Accuracy: 0.91

The accuracy score in the new model actually improves using the new model. Now, I will try a new model that does not have any insignificant predictors.

# Creating New Model

features_adj_1 = ["Diff_Str", "Diff_Td", "Diff_Head", "Diff_Body", "Diff_Leg", 
            "Diff_Ground", "Diff_Ctrl"]

ud_X_train_adj.drop(["Diff_Sub", "Diff_Clinch", "Diff_Kd"], axis = 1, inplace = True)

scaler = StandardScaler()
ud_X_train_adj_scaled = scaler.fit_transform(ud_X_train_adj)
ud_X_train_adj_with_intercept = add_constant(ud_X_train_adj_scaled)

ud_logit_model_adj = sm.Logit(ud_y_train, ud_X_train_adj_with_intercept)
ud_result_adj_1 = ud_logit_model_adj.fit()
print(ud_result_adj_1.summary(xname=["const"] + features_adj_1))

Optimization terminated successfully.
         Current function value: 0.234980
         Iterations 8
                           Logit Regression Results                           
==============================================================================
Dep. Variable:            Fighter_1_W   No. Observations:                 1778
Model:                          Logit   Df Residuals:                     1770
Method:                           MLE   Df Model:                            7
Date:                Tue, 12 Dec 2023   Pseudo R-squ.:                  0.6404
Time:                        13:02:46   Log-Likelihood:                -417.79
converged:                       True   LL-Null:                       -1161.7
Covariance Type:            nonrobust   LLR p-value:                     0.000
===============================================================================
                  coef    std err          z      P>|z|      [0.025      0.975]
-------------------------------------------------------------------------------
const           1.2145      0.145      8.390      0.000       0.931       1.498
Diff_Str        0.5227      0.160      3.277      0.001       0.210       0.835
Diff_Td         0.2160      0.104      2.083      0.037       0.013       0.419
Diff_Head       2.8266      0.198     14.268      0.000       2.438       3.215
Diff_Body       0.8029      0.111      7.216      0.000       0.585       1.021
Diff_Leg        0.7413      0.097      7.611      0.000       0.550       0.932
Diff_Ground     0.7550      0.250      3.025      0.002       0.266       1.244
Diff_Ctrl       1.7921      0.148     12.138      0.000       1.503       2.082
===============================================================================

# Finding Accuracy

ud_X_test_adj.drop(["Diff_Sub", "Diff_Clinch", "Diff_Kd"], axis = 1, inplace = True)

ud_X_test_scaled_adj = scaler.transform(ud_X_test_adj)  
ud_X_test_with_intercept_adj = add_constant(ud_X_test_scaled_adj)

ud_y_pred = ud_result_adj_1.predict(ud_X_test_with_intercept_adj)

ud_y_pred_binary = (ud_y_pred > 0.5).astype(int)

accuracy_adj_2 = (ud_y_pred_binary == ud_y_test).mean()
print(f"Accuracy: {accuracy_adj_2:.2f}")

Accuracy: 0.91

The accuracy remains the same, but the model only contains significant

Split Decision Model

# Creating Subset of Data

sd_train = train[train["Method"] == "S-DEC"]
sd_test = test[test["Method"] == "S-DEC"]

sd_X_train = sd_train[["Diff_Kd", "Diff_Str", "Diff_Td", "Diff_Sub", "Diff_Head", "Diff_Body", "Diff_Leg", 
               "Diff_Clinch", "Diff_Ground", "Diff_Ctrl"]]
sd_y_train = sd_train["Fighter_1_W"]
sd_X_test = sd_test[["Diff_Kd", "Diff_Str", "Diff_Td", "Diff_Sub", "Diff_Head", "Diff_Body", "Diff_Leg", 
               "Diff_Clinch", "Diff_Ground", "Diff_Ctrl"]]
sd_y_test = sd_test["Fighter_1_W"]

# Creating Model

features = ["Diff_Kd", "Diff_Str", "Diff_Td", "Diff_Sub", "Diff_Head", "Diff_Body", "Diff_Leg", 
            "Diff_Clinch", "Diff_Ground", "Diff_Ctrl"]
scaler = StandardScaler()
sd_X_train_scaled = scaler.fit_transform(sd_X_train)
sd_X_train_with_intercept = add_constant(sd_X_train_scaled)

sd_logit_model = sm.Logit(sd_y_train, sd_X_train_with_intercept)
sd_result = sd_logit_model.fit()
print(sd_result.summary(xname=["const"] + features))

Optimization terminated successfully.
         Current function value: 0.615246
         Iterations 5
                           Logit Regression Results                           
==============================================================================
Dep. Variable:            Fighter_1_W   No. Observations:                  477
Model:                          Logit   Df Residuals:                      466
Method:                           MLE   Df Model:                           10
Date:                Tue, 12 Dec 2023   Pseudo R-squ.:                 0.09707
Time:                        13:02:56   Log-Likelihood:                -293.47
converged:                       True   LL-Null:                       -325.02
Covariance Type:            nonrobust   LLR p-value:                 9.337e-10
===============================================================================
                  coef    std err          z      P>|z|      [0.025      0.975]
-------------------------------------------------------------------------------
const           0.3484      0.100      3.489      0.000       0.153       0.544
Diff_Kd         0.1818      0.105      1.732      0.083      -0.024       0.387
Diff_Str        0.0282      0.118      0.239      0.811      -0.203       0.259
Diff_Td         0.1484      0.109      1.367      0.172      -0.064       0.361
Diff_Sub        0.0334      0.104      0.320      0.749      -0.171       0.238
Diff_Head       0.4809      0.127      3.787      0.000       0.232       0.730
Diff_Body       0.1866      0.116      1.610      0.107      -0.041       0.414
Diff_Leg        0.1800      0.110      1.642      0.101      -0.035       0.395
Diff_Clinch     0.2843      0.122      2.333      0.020       0.045       0.523
Diff_Ground     0.1526      0.116      1.316      0.188      -0.075       0.380
Diff_Ctrl       0.4915      0.115      4.283      0.000       0.267       0.716
===============================================================================

The table indicates that the Pseudo R-Squared Value is much lower than the above models. Only 3 predictors are significant in the model above: Head Strikes Differential, Clinch Strikes Differential, and Control Time Differential.

# Finding Accuracy

sd_X_test_scaled = scaler.transform(sd_X_test)  # Use the same scaler as on the training set
sd_X_test_with_intercept = add_constant(sd_X_test_scaled)

# Predict on the test set
sd_y_pred = sd_result.predict(sd_X_test_with_intercept)

# Convert predicted probabilities to binary predictions (0 or 1)
sd_y_pred_binary = (sd_y_pred > 0.5).astype(int)

# Calculate accuracy
accuracy = (sd_y_pred_binary == sd_y_test).mean()
print(f"Accuracy: {accuracy:.2f}")

Accuracy: 0.61

An accuracy score of 0.61 is lower than desired, as the goal score was 0.67 or higher (as 2 of the 3 judges agree on the winner). To address this, the insignificant predictors are adjusted as they were in the other models.

# Creating Dummy Variables

sd_X_train_adj = sd_X_train.copy()

third_quartile_value = data["Diff_Str"].quantile(0.90)
sd_X_train_adj.loc[:, "Diff_Str"] = sd_X_train_adj["Diff_Str"].apply(lambda x: 1 if x > third_quartile_value else 0)

third_quartile_value = data["Diff_Td"].quantile(0.90)
sd_X_train_adj.loc[:, "Diff_Td"] = sd_X_train_adj["Diff_Td"].apply(lambda x: 1 if x > third_quartile_value else 0)

third_quartile_value = data["Diff_Sub"].quantile(0.90)
sd_X_train_adj.loc[:, "Diff_Sub"] = sd_X_train_adj["Diff_Sub"].apply(lambda x: 1 if x > third_quartile_value else 0)

third_quartile_value = data["Diff_Body"].quantile(0.90)
sd_X_train_adj.loc[:, "Diff_Body"] = sd_X_train_adj["Diff_Body"].apply(lambda x: 1 if x > third_quartile_value else 0)

third_quartile_value = data["Diff_Leg"].quantile(0.90)
sd_X_train_adj.loc[:, "Diff_Leg"] = sd_X_train_adj["Diff_Leg"].apply(lambda x: 1 if x > third_quartile_value else 0)

third_quartile_value = data["Diff_Ground"].quantile(0.90)
sd_X_train_adj.loc[:, "Diff_Ground"] = sd_X_train_adj["Diff_Ground"].apply(lambda x: 1 if x > third_quartile_value else 0)

sd_X_test_adj = sd_X_test.copy()

third_quartile_value = data["Diff_Str"].quantile(0.90)
sd_X_test_adj.loc[:, "Diff_Str"] = sd_X_test_adj["Diff_Str"].apply(lambda x: 1 if x > third_quartile_value else 0)

third_quartile_value = data["Diff_Td"].quantile(0.90)
sd_X_test_adj.loc[:, "Diff_Td"] = sd_X_test_adj["Diff_Td"].apply(lambda x: 1 if x > third_quartile_value else 0)

third_quartile_value = data["Diff_Sub"].quantile(0.90)
sd_X_test_adj.loc[:, "Diff_Sub"] = sd_X_test_adj["Diff_Sub"].apply(lambda x: 1 if x > third_quartile_value else 0)

third_quartile_value = data["Diff_Body"].quantile(0.90)
sd_X_test_adj.loc[:, "Diff_Body"] = sd_X_test_adj["Diff_Body"].apply(lambda x: 1 if x > third_quartile_value else 0)

third_quartile_value = data["Diff_Leg"].quantile(0.90)
sd_X_test_adj.loc[:, "Diff_Leg"] = sd_X_test_adj["Diff_Leg"].apply(lambda x: 1 if x > third_quartile_value else 0)

third_quartile_value = data["Diff_Ground"].quantile(0.90)
sd_X_test_adj.loc[:, "Diff_Ground"] = sd_X_test_adj["Diff_Ground"].apply(lambda x: 1 if x > third_quartile_value else 0)

# Creating New Model

features = ["Diff_Kd_Cat", "Diff_Str_Cat", "Diff_Td_Cat", "Diff_Sub_Cat", "Diff_Head", "Diff_Body_Cat", "Diff_Leg_Cat", 
            "Diff_Clinch", "Diff_Ground_Cat", "Diff_Ctrl"]
scaler = StandardScaler()
sd_X_train_adj_scaled = scaler.fit_transform(sd_X_train_adj)
sd_X_train_adj_with_intercept = add_constant(sd_X_train_adj_scaled)

sd_adj_logit_model = sm.Logit(sd_y_train, sd_X_train_adj_with_intercept)
sd_result_adj = sd_adj_logit_model.fit()
print(sd_result_adj.summary(xname=["const"] + features))

Optimization terminated successfully.
         Current function value: 0.619419
         Iterations 5
                           Logit Regression Results                           
==============================================================================
Dep. Variable:            Fighter_1_W   No. Observations:                  477
Model:                          Logit   Df Residuals:                      466
Method:                           MLE   Df Model:                           10
Date:                Tue, 12 Dec 2023   Pseudo R-squ.:                 0.09095
Time:                        13:03:04   Log-Likelihood:                -295.46
converged:                       True   LL-Null:                       -325.02
Covariance Type:            nonrobust   LLR p-value:                 5.313e-09
===================================================================================
                      coef    std err          z      P>|z|      [0.025      0.975]
-----------------------------------------------------------------------------------
const               0.3509      0.099      3.533      0.000       0.156       0.546
Diff_Kd_Cat         0.1576      0.103      1.524      0.127      -0.045       0.360
Diff_Str_Cat       -0.0407      0.099     -0.410      0.682      -0.235       0.154
Diff_Td_Cat        -0.0416      0.100     -0.416      0.678      -0.237       0.154
Diff_Sub_Cat        0.1777      0.109      1.635      0.102      -0.035       0.391
Diff_Head           0.4444      0.113      3.941      0.000       0.223       0.665
Diff_Body_Cat       0.1346      0.105      1.279      0.201      -0.072       0.341
Diff_Leg_Cat        0.0788      0.104      0.759      0.448      -0.125       0.282
Diff_Clinch         0.3669      0.114      3.206      0.001       0.143       0.591
Diff_Ground_Cat    -0.0500      0.112     -0.446      0.656      -0.270       0.170
Diff_Ctrl           0.4914      0.108      4.553      0.000       0.280       0.703
===================================================================================

# Finding Accuracy

sd_X_test_scaled_adj = scaler.transform(sd_X_test_adj)  # Use the same scaler as on the training set
sd_X_test_with_intercept_adj = add_constant(sd_X_test_scaled_adj)

# Predict on the test set
sd_y_pred = sd_result_adj.predict(sd_X_test_with_intercept_adj)

# Convert predicted probabilities to binary predictions (0 or 1)
sd_y_pred_binary = (sd_y_pred > 0.5).astype(int)

# Calculate accuracy
accuracy = (sd_y_pred_binary == sd_y_test).mean()
print(f"Accuracy: {accuracy:.2f}")

Accuracy: 0.63

The accuracy score increases when the model is adjusted. What if all insignificant predictors are simply dropped?

# Creating New Model

features_sd = ["Diff_Head", "Diff_Clinch", "Diff_Ctrl"]

sd_X_train_adj.drop(["Diff_Kd", "Diff_Str", "Diff_Td", "Diff_Sub", "Diff_Body", "Diff_Leg", "Diff_Ground"], axis = 1, inplace = True)
scaler = StandardScaler()
sd_X_train_adj_scaled = scaler.fit_transform(sd_X_train_adj)
sd_X_train_adj_with_intercept = add_constant(sd_X_train_adj_scaled)

sd_adj_logit_model = sm.Logit(sd_y_train, sd_X_train_adj_with_intercept)
sd_result_adj = sd_adj_logit_model.fit()
print(sd_result_adj.summary(xname=["const"] + features_sd))

Optimization terminated successfully.
         Current function value: 0.627974
         Iterations 5
                           Logit Regression Results                           
==============================================================================
Dep. Variable:            Fighter_1_W   No. Observations:                  477
Model:                          Logit   Df Residuals:                      473
Method:                           MLE   Df Model:                            3
Date:                Tue, 12 Dec 2023   Pseudo R-squ.:                 0.07839
Time:                        13:03:12   Log-Likelihood:                -299.54
converged:                       True   LL-Null:                       -325.02
Covariance Type:            nonrobust   LLR p-value:                 4.991e-11
===============================================================================
                  coef    std err          z      P>|z|      [0.025      0.975]
-------------------------------------------------------------------------------
const           0.3420      0.098      3.483      0.000       0.150       0.534
Diff_Head       0.4052      0.107      3.774      0.000       0.195       0.616
Diff_Clinch     0.3990      0.111      3.602      0.000       0.182       0.616
Diff_Ctrl       0.4997      0.106      4.727      0.000       0.293       0.707
===============================================================================

# Finding Accuracy

sd_X_test_adj.drop(["Diff_Kd", "Diff_Str", "Diff_Td", "Diff_Sub", "Diff_Body", "Diff_Leg", "Diff_Ground"], axis = 1, inplace = True)

sd_X_test_scaled_adj = scaler.transform(sd_X_test_adj)  # Use the same scaler as on the training set
sd_X_test_with_intercept_adj = add_constant(sd_X_test_scaled_adj)

# Predict on the test set
sd_y_pred = sd_result_adj.predict(sd_X_test_with_intercept_adj)

# Convert predicted probabilities to binary predictions (0 or 1)
sd_y_pred_binary = (sd_y_pred > 0.5).astype(int)

# Calculate accuracy
accuracy_adj_3 = (sd_y_pred_binary == sd_y_test).mean()
print(f"Accuracy: {accuracy_adj_3:.2f}")

Accuracy: 0.62

Although the accuracy in the model is slightly lower than the model before, it only contains significant predictors, and so this is the model that would be selected.

Final Models

Below are the best models as associated accuracy scores.

Model for All Fights

# Model
print(result_adj_1.summary(xname=["const"] + features_adj))

                           Logit Regression Results                           
==============================================================================
Dep. Variable:            Fighter_1_W   No. Observations:                 2312
Model:                          Logit   Df Residuals:                     2303
Method:                           MLE   Df Model:                            8
Date:                Tue, 12 Dec 2023   Pseudo R-squ.:                  0.4750
Time:                        13:03:16   Log-Likelihood:                -801.38
converged:                       True   LL-Null:                       -1526.4
Covariance Type:            nonrobust   LLR p-value:                8.244e-308
===============================================================================
                  coef    std err          z      P>|z|      [0.025      0.975]
-------------------------------------------------------------------------------
const           0.8637      0.083     10.448      0.000       0.702       1.026
Diff_Kd         0.1662      0.066      2.516      0.012       0.037       0.296
Diff_Str        0.3665      0.103      3.552      0.000       0.164       0.569
Diff_Td         0.2020      0.072      2.815      0.005       0.061       0.343
Diff_Head       1.9475      0.118     16.515      0.000       1.716       2.179
Diff_Body       0.6177      0.075      8.217      0.000       0.470       0.765
Diff_Leg        0.5936      0.069      8.586      0.000       0.458       0.729
Diff_Ground     0.3152      0.145      2.174      0.030       0.031       0.599
Diff_Ctrl       1.3941      0.099     14.153      0.000       1.201       1.587
===============================================================================

# Accuracy for Model
print(f"Accuracy: {accuracy_adj_1:.2f}")

Accuracy: 0.86

Model for Unanimous Decisions

# Model
print(ud_result_adj_1.summary(xname=["const"] + features_adj_1))

                           Logit Regression Results                           
==============================================================================
Dep. Variable:            Fighter_1_W   No. Observations:                 1778
Model:                          Logit   Df Residuals:                     1770
Method:                           MLE   Df Model:                            7
Date:                Tue, 12 Dec 2023   Pseudo R-squ.:                  0.6404
Time:                        13:03:20   Log-Likelihood:                -417.79
converged:                       True   LL-Null:                       -1161.7
Covariance Type:            nonrobust   LLR p-value:                     0.000
===============================================================================
                  coef    std err          z      P>|z|      [0.025      0.975]
-------------------------------------------------------------------------------
const           1.2145      0.145      8.390      0.000       0.931       1.498
Diff_Str        0.5227      0.160      3.277      0.001       0.210       0.835
Diff_Td         0.2160      0.104      2.083      0.037       0.013       0.419
Diff_Head       2.8266      0.198     14.268      0.000       2.438       3.215
Diff_Body       0.8029      0.111      7.216      0.000       0.585       1.021
Diff_Leg        0.7413      0.097      7.611      0.000       0.550       0.932
Diff_Ground     0.7550      0.250      3.025      0.002       0.266       1.244
Diff_Ctrl       1.7921      0.148     12.138      0.000       1.503       2.082
===============================================================================

# Accuracy for Model
print(f"Accuracy: {accuracy_adj_2:.2f}")

Accuracy: 0.91

Model for Split Decisions

# Model
print(sd_result_adj.summary(xname=["const"] + features_sd))

                           Logit Regression Results                           
==============================================================================
Dep. Variable:            Fighter_1_W   No. Observations:                  477
Model:                          Logit   Df Residuals:                      473
Method:                           MLE   Df Model:                            3
Date:                Tue, 12 Dec 2023   Pseudo R-squ.:                 0.07839
Time:                        13:03:25   Log-Likelihood:                -299.54
converged:                       True   LL-Null:                       -325.02
Covariance Type:            nonrobust   LLR p-value:                 4.991e-11
===============================================================================
                  coef    std err          z      P>|z|      [0.025      0.975]
-------------------------------------------------------------------------------
const           0.3420      0.098      3.483      0.000       0.150       0.534
Diff_Head       0.4052      0.107      3.774      0.000       0.195       0.616
Diff_Clinch     0.3990      0.111      3.602      0.000       0.182       0.616
Diff_Ctrl       0.4997      0.106      4.727      0.000       0.293       0.707
===============================================================================

# Accuracy for Model
print(f"Accuracy: {accuracy_adj_3:.2f}")

Accuracy: 0.62

Conclusions

Takeaways from the Logistic Regression Models

Based on the first model, which does not take into account the type of outcome, one can determine that Significant Strike Differential, Takedown Differential, Strikes to the Head Differential, Strikes to the Body Differential, Leg Kicks Differential, Control Time Differential, Ground Strikes Differential, and Knockdown Differential are the most important features when predicting the winner of a fight, with Strikes to the Head Differential and Control Time Differential having the largest impact on the judges' scorecards. This is not terribly surprising, as these two features are perhaps the best indicators of effective striking and effective grappling, which are the two most important standards by which judges are supposed to score fights. The next two important features are Body Strike Differential and Leg Kicks Differential, respectively, which are also indicators of effective striking. However, the main idea is that, since there are multiple significant predictors, a fighter can find success in a fight by gaining advantages in a variety of areas. That being said, certain parts of their output, namely Strikes to the Head Differential and Control Time Differential, should be prioritized.

The Unanimous Decision Model offers similar insights. When all judges agree on a winner, the model suggests that there are multiple different areas in the fight that offer significant insight on who won. Interestingly, Knockdown Differential is not a significant predictor when modeling on this subset of the data. All this being said, Strikes to the Head Differential and Control Time Differential are still the most impactful predictors.

What is perhaps most surprising are the takeaways from the Split Decision Model. Based on the EDA, there were doubts as to whether there would be any significant predictors from the features in the data. While the number of significant predictors in the model are lower than the other two models, Strikes to the Head Differential and Control Time Differential were significant and still have the largest coefficients. This would indicate that in a close fight, the features most strongly related to the most important judging criterion play the most significant role in a judge's mind when discerning the winner in a close fight. This sort of information would be particularly useful to fighters and coaches, as if the fight is close and the fight is coming to an end, a strategy could be formed so as to gain advantages in the competition.