Research on Applying Machine Learning to Improve Player Valuation For Scouting in a Football Team

This project addresses the challenge of inefficient and subjective football scouting by developing a data-driven machine learning system to improve player valuation and recruitment decisions. It optimizes manual, intuition-based evaluation with a scalable approach that analyses player performance data, identifies similar players, and evaluates team fit through chemistry metrics. The goal is to help clubs make more consistent, accurate, and efficient scouting decisions in an increasingly competitive and data-rich football environment.

The proposed machine learning system improves player scouting by producing more accurate and meaningful player comparisons and recommendations. The VAE-Gamma model significantly outperforms traditional methods in clustering player data, leading to better similarity matching. The chemistry metrics, particularly Joint Offensive Impact (JOI), align well with real-world transfer decisions and benchmark rankings, achieving solid recommendation performance (e.g., consistent shortlist rankings and a Hit@10 rate of 0.45).