Optimization and Machine Learning techniques: An application to the Fantasy Premier League

Fantasy sports have become a fertile ground for applying advanced analytics and optimization tech- niques. This thesis explores the development of a machine learning and operations research pipeline for optimizing team selection in the Fantasy Premier League (FPL). Using historical data, we first train predictive models to estimate player performances. These predictions are then embedded in Mixed Integer Programming (MIP) formulations to select optimal squads under realistic FPL constraints. We further refine our approach by incorporating weekly updates, market dynamics, and captain selection strategies. The results demonstrate that combining predictive analytics with mathematical optimiza- tion significantly improves final season scores compared to heuristic or greedy approaches. This study showcases the synergy between data-driven modeling and decision-making under uncertainty, with potential applications extending beyond fantasy sports.