Machine Learning in Fantasy Premier League

Abstract

The world of fantasy sports has witnessed a surge in popularity, with Fantasy Premier League (FPL) standing out as a leader in the field. This diploma thesis focuses on the application of machine learning techniques in Fantasy Premier League (FPL) to enhance team performance and decision-making. The research objectives were threefold: to develop machine learning models for predicting Expected Value (EV), to generate optimal moves based on the predicted EV values, and to contribute to the field of FPL analytics by advancing the application of machine learning techniques.

The growing popularity of FPL (over 11.4 million teams in the 2022/23 season) has increased the demand for data-driven strategies across the community to gain a competitive edge. While previous efforts by the FPL community have explored similar concepts, our study presents the first open-source, full-scale FPL model utilizing machine learning techniques.

The results of our research demonstrate the superiority of our model compared to most of the other existing models. Through rigorous evaluation using various metrics (MAE, RMSE, R² score), our models consistently outperformed almost all competitors in predicting EV. Furthermore, a noteworthy achievement was the top-ranking performance of the team managed by our model in the GW24 league, surpassing over 25,000 competing teams by the end of the season.

By successfully integrating machine learning and optimization algorithms, our project provides FPL managers with actionable recommendations for team selection, transfers, and strategic decision-making. This research significantly contributes to the field of FPL Analytics by offering an innovative solution that enhances team performance and provides valuable insights for managers seeking to improve their FPL results.