Data-Driven Football Scouting: Cosine-Based Player Similarity and Random Forest Transfer Prediction in EA Sports FC 24

Abstract

We introduce a data-driven football scouting framework utilizing the EA Sports FC 24 player dataset. The system comprises two main modules: (1) Player Similarity Search, where each player is represented by a normalized skill-attribute vector and the k most similar players are retrieved using cosine similarity, and (2) Transfer Likelihood Prediction, a Random Forest classifier that estimates the probability a player will transfer in the next window based on contract and performance features. The dataset contains roughly 18,000 players with about 70 attributes each, and we benchmark our model against a logistic regression baseline. Our key contributions include the integration of unsupervised similarity search with supervised transfer prediction and the addition of interpretable visualizations. Specifically, we present multi-attribute radar charts and scatter plots that help scouts understand why players are recommended. This hybrid approach automatically flags similarly skilled players and quantifies their expected transfer probability, offering a practical decision-support tool for talent identification.