Football results prediction and machine learning techniques

In the past, machine learning techniques used to predict the outcome of professional team-based sports matches have used the number of points or goals scored as the primary metric for performance evaluation in their prediction models. However, this approach is considered outdated by industry statisticians. The final outcome of each match can fluctuate wildly from the expected outcome based on events and changes of circumstances occurring within the games. The aim of this project is to compare and contrast the effectiveness and performance of various machine learning models when predicting the outcome of football matches in the English Premier League, both to each other and other benchmarks, including bookmakers' models and random chance. In this research, the 'expected goals' metric was explored as the base of the machine learning algorithms instead of the traditional 'goals scored' metric. This was used to build a Poisson distribution probabilistic classifier to predict the results of matches in the future, achieving an accuracy of 52.3% with regard to matches that occurred during the 2020-2021 Premier League season.