Revisiting Expected Possession Value in Football: Introducing a Benchmark, U-Net Architecture, and Reward and Risk for Passes

TL;DR

This work tackles the evaluation and enhancement of Expected Possession Value (EPV) models in football by introducing the OJN-Pass-EPV benchmark and a U‑Net–based EPV model that incorporates ball height and a reward/risk decomposition for passes. It demonstrates that relative EPV judgments can be reliably predicted across Eredivisie and World Cup data, achieving 78% accuracy on the benchmark and improving interpretability through separate reward and risk components. The approach addresses replication challenges in prior work and emphasizes cross-competition robustness, calibration, and domain-specific features. Overall, the paper provides a standardized evaluation framework and a more accurate, interpretable EPV model to inform tactical decisions and performance analysis.

Abstract

This paper introduces the first Expected Possession Value (EPV) benchmark and a new and improved EPV model for football. Through the introduction of the OJN-Pass-EPV benchmark, we present a novel method to quantitatively assess the quality of EPV models by using pairs of game states with given relative EPVs. Next, we attempt to replicate the results of Fernández et al. (2021) using a dataset containing Dutch Eredivisie and World Cup matches. Following our failure to do so, we propose a new architecture based on U-net-type convolutional neural networks, achieving good results in model loss and Expected Calibration Error. Finally, we present an improved pass model that incorporates ball height and contains a new dual-component pass value model that analyzes reward and risk. The resulting EPV model correctly identifies the higher value state in 78% of the game state pairs in the OJN-Pass-EPV benchmark, demonstrating its ability to accurately assess goal-scoring potential. Our findings can help assess the quality of EPV models, improve EPV predictions, help assess potential reward and risk of passing decisions, and improve player and team performance.

Figures (3)

• Figure 1: Comparative visualization of pass likelihood based on ball height. In both scenarios, the red team (red dots) is in possession, with the ball carrier highlighted in yellow and the ball depicted in violet. The opposing team is represented in blue (blue dots). The color intensity on the pitch indicates the likelihood of the pass reaching a location - the darker the shade, the higher the probability. Figure \ref{['fig:ball_height_aerial']} shows how the model recognizes that aerial passes can be made over opponents, while also showing increased uncertainty about the pass destination compared to the ground pass in Figure \ref{['fig:ball_height_ground']}.
• Figure 2: Game state showing a pass scenario. The ball carrier, marked with a yellow dot and belonging to the red team, is making a pass to the target location marked by "+".
• Figure 3: Analysis of pass decisions with OJN-EPV (based on the output in Definition \ref{['def:output']}). This figure shows a player's actual pass (marked by "+") against the optimal pass location (circled) as identified by OJN-EPV. The comparison highlights potential areas for decision-making refinement, illustrating how OJN-EPV can assist in identifying more dangerous pass options.

Theorems & Definitions (1)

• Definition 1