AI in Bundesliga Match Analysis – Expected Possession Value (EPV) vs. Expected Goals (xG) to Predict Match Outcomes in Soccer

Leander Forcher^1*Leon Forcher²Alexander Woll¹Stefan Altmann¹

• ¹Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany
• ²TSG 1899 Hoffenheim, Zuzenhausen, Germany

With an increasing number of key performance indicators (KPIs) in soccer analytics, it is key to identify the most valuable KPIs. One approach to define a KPI's value is to assess its ability to predict match outcomes and future performance. Therefore, this study aims to compare the effectiveness of expected goals (xG) and expected possession value (EPV) in predicting match outcomes in both pre-match and post-match scenarios. Event and tracking data of two Bundesliga seasons (2022/23 & 2023/24) were used to develop four distinct match outcome prediction approaches: xG & EPV pre-match (using features including the last three match performances of teams & contextual factors) and xG & EPV post-match (using xG and EPV performances of the played match). The xG post-match prediction showed the best performance in predicting match outcomes (xG post-match: RPS=0.148, Accuracy=0.656; EPV post-match: RPS=0.191, Accuracy=0.596). In pre-match scenarios EPV showed higher prediction performance (RPS=0.194, Accuracy=0.583) compared to xG (RPS=0.199, Accuracy=0.556). Accordingly, xG holds more valuable performance information on the offensive performance of a team in post-match scenarios. In contrast, the EPV pre-match prediction showed powerful results in predicting future match outcomes and thereby showcased the predictiveness of EPV.