EPL Match Prediction via Random Forest Classification
This repository moves beyond manual weights to implement a Random Forest Classifier (RFC). By training on over 500 Premier League matches across two seasons, the model learns the non-linear relationships between team stats and the final result.
How it Works: The model consists of an ensemble of 200 Decision Trees. Each tree analyzes a random subset of features (like Goal Difference, Red Cards, or Rest Days) to vote on the most likely outcome.
Data Science Highlights:
-
Feature Engineering: Combines 14 distinct features per match, including defense ratings and efficiency metrics.
-
Model Persistence: Uses joblib to save the trained "brain" (.pkl), allowing for instant inference without retraining.
-
Explainable AI: Includes scripts to visualize Feature Importance, showing which stats actually drive wins in the Premier League.
-
Multi-Season Training: Handles the "Promoted Team Problem" by using league-average baselines for teams newly arrived from the Championship.
Visualizing the Brain:
The project generates a feature importance graph to show which metrics the model "valued" most during training.
Tech Stack: Python
Scikit-Learn (Random Forest, Label Encoding)
Pandas (Data manipulation)
Matplotlib (Visualization)
Joblib (Model Serialization)
Data was collected from https://www.football-data.co.uk/englandm.php