Soccer Player Ratings
This project proposed a Regularized Adjusted Expected Goal Plus-Minus Model (RAxGPM) in soccer to evaluate players’ performance, which incorporates priors generated from box-score data into a regularized regression framework and uses expected goals differential per 90 as the response variable. In specific, player-level box-score statistics regressed on FIFA ratings are used to construct prior ratings for each player, and stint-by-stint data are created to fit the RAPM framework. The results have shown that the RAxGPM model with prior created from box-score statistics has the better descriptive and predictive ability, as the model has the highest prediction accuracy among the models we have compared.
Joint work with Edvin Tran Hoac, and Phong Hoang.