EPL Match Outcome Prediction

Machine learning model for predicting English Premier League football match outcomes

Course: COMP0036 - Machine Learning for Natural and Computational Sciences Group: K Academic Year: 2025/26

---

Overview

This project develops machine learning models to predict English Premier League (EPL) football match outcomes. Given historical match data from 2000-2024, we predict whether matches will result in a home win, draw, or away win.

Task

Predict the outcomes of 10 EPL matches scheduled for 31 January 2026.

Target Variable: FTR (Full Time Result)

• H - Home Win
• D - Draw
• A - Away Win

Data

Training Set: ~9,600 EPL matches (2000-2024)

• Match details: Date, teams, referee
• Goals: Full-time and half-time scores
• Statistics: Shots, shots on target, corners, fouls
• Cards: Yellow and red cards

Test Set: 10 matches on 31 January 2026

Methodology

The project focuses on:

1. Exploratory Data Analysis: Understanding patterns in historical match data
2. Feature Engineering: Creating predictive features from raw statistics (team form, head-to-head records, home advantage, etc.)
3. Model Development: Training and comparing multiple ML algorithms
4. Model Evaluation: Validation using appropriate train/test splits

External data sources may be used to enhance predictions (player stats, manager info, etc.), excluding betting odds.

Benchmark

Professional bookmakers achieve ~53% accuracy in match prediction. This serves as the performance benchmark for the task.

Note: Assessment prioritizes methodology, understanding, and creativity over absolute accuracy.

Deliverables

• Report (PDF): Methodology, analysis, and findings
• Notebook (.ipynb): Implementation with documentation
• Predictions (CSV): Test set predictions