Predicting Football Matches Outcomes Using Machine Learning Approaches

Abstract

This thesis delves into the intersection of machine learning and football match prediction, exploring the evolving landscape where data-driven methodologies are transforming the sport. With a focus on the English Premier League, we employ a meticulous blend of data collection, preprocessing, and feature engineering to develop robust predictive models. Our investigation spans both multiclass and binary classifications, utilizing popular algorithms such as Linear Support Vector Classifier, Logistic Regression, Random Forest, Extreme Gradient Boosting, and Light Gradient Boosting Machine. The results reveal the Light Gradient Boosting Machine as the most effective model, achieving a notable accuracy of 73% in multiclass prediction and 80.44% in binary classification. Profit simulation based on betting odds further showcases the model's potential, yielding an average profit gain of 72.2% across seasons on distributive betting and reaches up to x186 of profit on cumulative betting while strategized with Kelly criterion. Feature importance analysis underscores the significance of team standings, win percentages, and historical performance metrics, shedding light on the intricate dynamics influencing football match outcomes.

Date of Award	2024
Original language	American English
Awarding Institution	HBKU College of Science and Engineering