The Whole Package: All-In-One Player Performance Metrics for Soccer

Abstract

Analytics is a growing part of modern soccer, with many teams utilizing a variety of quantitative methods to help improve their odds of winning. In this paper, I develop two novel all-in-one metrics for player performance in soccer through applying supervised learning methods. One metric is a goal probability added (GPA) metric, while the other is a win probability added (WPA) metric. I also contribute one of the first efforts in the literature to formalize a unified meta-evaluation procedure for assessing all-in-one player performance metrics. Running the meta-evaluation procedure, I find that both GPA and WPA perform well across the criteria of ease of understanding, granularity, modeling performance, and feature-level stability.