Predicting Football Match Outcomes with eXplainable Machine Learning and the Kelly Index

In this work, a machine learning approach is developed for predicting the outcomes of football matches. The novelty of this research lies in the utilisation of the Kelly Index to first classify matches into categories where each one denotes the different levels of predictive difficulty. Classification models using a wide suite of algorithms were developed for each category of matches in order to determine the efficacy of the approach. In conjunction to this, a set of previously unexplored features were engineering including Elo-based variables. The dataset originated from the Premier League match data covering the 2019-2021 seasons. The findings indicate that the process of decomposing the predictive problem into sub-tasks was effective and produced competitive results with prior works, while the ensemble-based methods were the most effective. The paper also devised an investment strategy in order to evaluate its effectiveness by benchmarking against bookmaker odds. An approach was developed that minimises risk by combining the Kelly Index with the predefined confidence thresholds of the predictive models. The experiments found that the proposed strategy can return a profit when following a conservative approach that focuses primarily on easy-to-predict matches where the predictive models display a high confidence level.