DS-Project-5-EPL-Win_Predictor ⚽

Created a model that predicts the chances of winning a premiere league football match for your favourite EPL team.

• Designed a model that predicts the chances of winning a football ⚽ match for your favourite premiere league team.
• Dataset - Scraped and fetched data from the resources provided below . The Dataset contains some important statistics for the last 12 seasons of the English Premier League.
• Model - The major aim of in this project is to build a predictor that predicts the chances of winning a game based on the previous statistical data by using classification algorithms and techniques.

Code and Resources Used

Python Version: 3.10.5
Packages: pandas, pandas-profiling, numpy, sklearn, matplotlib, seaborn
For Web Framework Requirements: pip install -r requirements.txt
Data Resources: The Dataset is originally sourced from http://www.football-data.co.uk/ but later I found out a Kaggle resource from which it was easier to fetch the data. https://www.kaggle.com/datasets/lumierebatalong/english-premiere-league-team-datasets

About the Dataset

• Barclay premier league is the best ⚽ league in the world.It consists of the best 20 English football clubs that have qualified to fight for the title.
• Every team plays every other team twice-once at home ground which is known as a Home game and once at the opponent's home ground which is termed as the Away game in a single calender year.
• The Kaggle Dataset consists of statistical data of the previous 12 seasons for 5 teams which have already won the title in the last 12 seasons namely Man City, Liverpool, Man United, Chelsea, Leicester with two outsiders Arsenal and Tottenham [Note: If your favourite team is not in the list, you can scrape the data from the source website linked above using libraries like BeautifulSoup or Selenium.]

Data Overview

The Features are defined as follows:

Div = League Division
Date = Match Date (dd/mm/yy)
HomeTeam = Home Team
AwayTeam = Away Team
FTHG = Full Time Home Team Goals
FTAG = Full Time Away Team Goals
FTR = Full Time Result (H=Home Win, D=Draw, A=Away Win)
HTHG = Half Time Home Team Goals
HTAG = Half Time Away Team Goals
HTR = Half Time Result (H=Home Win, D=Draw, A=Away Win)
Referee = Match Referee
HS = Home Team Shots
AS = Away Team Shots
HST = Home Team Shots on Target
AST = Away Team Shots on Target
HC = Home Team Corners
AC = Away Team Corners
HF = Home Team Fouls Committed
AF = Away Team Fouls Committed
HY = Home Team Yellow Cards
AY = Away Team Yellow Cards
HR = Home Team Red Cards
AR = Away Team Red Cards

Data Preprocessing and Cleaning

• Fortunately, we dont have any null values or missing values. We dont have to use an Imputer.
• I remove unneceassry attributes such as Div ,Date,Referee.
• Since I want to calculate the total number of wins,losses and draws for my team , I define a function that considers the HomeTeam,AwayTeam column and based on the 'FTR' which is the full time result, it returns me a list of wins,losses and draws for Man City. I add this list as a column which will later be the label for my model as I need to predict the end result of a game.
• Then I drop the 'FTR' column as I dont require it anymore.
• Finally, I manipulate the 'str' datatype features as an 'int' datatype. This is because a classifier gives us better results on 'int' datatype.

This is how the data looks like after Preprocessing it:

EDA - Exploratory data analysis

• I looked at the distributions of the data and the value counts for thenumber of wins ,losses and draws.
• Statistical insights on this dataset using the pandas describe() function allows us to really dive into analytics of football and the performance of a club in football games. I am not really diving into the analytics part as our job is to focus on the ML model.
• I also plotted a heatmap for the correlations between the features and label.
• Besides this, I have also used the pandas-profiling library to generate full detailed EDA report on the dataset. Please find it in the repo for reference.

Model Building

• I split the data into train and tests sets with a test size of 15%.
• After splitting, I stratefied shuffled the datasets to make sure there are similar ratios of the 'Label' attribute in both the testing and training datasets.
• Finally, I split the data into features and label

As I am creating a Win predictor , I manipulate my labels to return True for Wins and False for losses and draws. This is my binary condition for my classifier.

Creating a Pipeline

This is just to automate all my processes.
I have added a StandardScaler for scaling my features and fit_transformed my features for the model.
[Note: You can add all processes to this pipeline such as Imputers, etc.]

Selecting and training a desired model for our predictor

I tried five different classifier models and evaluated them by calculating the Precision,Recall and F1-Score. I believe accuracy is not the best way to evaluate classifiers, therefore I calculated the F-1 scores for an accurate evaluation for our model.

Although Before calculating the precision,recall and F1-score, I had to first compile the confusion matrix which gives us an idea of how many predictions were right and wrong.

The confusion matrix tell us that:

• 57 False outcomes were predicted as True outcomes
  
• 78 True outcomes were preicted as False outcomes
  

The five different models I tried are:

1. LogisticRegression - This is the first approach i take to every classying problem as it is efficient. The feedback was average for this particular dataset.
2. GaussianNB - Naive Bayes classifiers are one of the best models for predictions. This gave me the best results.
3. RandomForestClassifier - Good with a lot of features. Performed the best on the training datasets with a very high f1-score of 91% but did not perform as good with the testing dataset.
4. DecisionTreeClassifier - Very similar performance as the RandomForestClassifier
5. Support Vector Machine Classifier (SVM) - In short this finds the best hyperplane and splits the data. Had a similar f1-score as the RandomForestClassifier

Model Performance

The Gaussian Naive Bayes Classifier performed the best on the test and validation sets.I found out the Precision,Recall and F1-Scores for all the models. This gives us a proper evaluation of the model. Below are the highlights of how the GaussianNB model performed :
Precision_score - 0.7241379310344828
Recall_score - 0.9333333333333333
F1_Score - 0.8155339805825242