Position-Aware Machine Learning Analysis of Football Shooting Performance

Abstract

Measuring individual shooting performance in professional football remains a challenging task because players differ considerably in tactical roles, playing time, and offensive responsibilities. Conventional indicators such as goals scored or total shots provide only partial assessments of shooting performance and often ignore positional expectations and playing-time normalization. Existing advanced metrics frequently rely on event-based or tracking data, limiting their accessibility for many practitioners. In this study, we propose the Position-Aware Shooting Impact Score (PASIS), an interpretable composite metric that integrates shooting efficiency, shooting accuracy, shot volume, positional role, and playing time into a unified performance indicator using publicly available player statistics. Rather than replacing the analytical PASIS formulation, machine learning is employed to investigate the nonlinear relationships embedded within the metric, compare the predictive capabilities of several regression algorithms, and identify the variables that most strongly influence shooting performance. A comparative evaluation was conducted using player-level shooting statistics from the English Premier League 2024--2025 season. Linear Regression, Ridge Regression, Lasso Regression, K-Nearest Neighbors, Gradient Boosting, and both baseline and hyperparameter-optimized Random Forest models were evaluated using coefficient of determination (R²), mean absolute error (MAE), and mean squared error (MSE). Model interpretability was further investigated using SHAP analysis. The experimental results indicate that ensemble learning methods outperform linear and distance-based models in approximating the nonlinear relationships underlying PASIS, with the tuned Random Forest achieving the highest predictive performance. The analysis further confirms that shooting efficiency, shooting accuracy, and shot volume are the dominant contributors to the metric. These findings demonstrate that PASIS provides an interpretable framework for analysing individual shooting performance while offering practical value for player evaluation, scouting, and football performance analysis.