Success Score: A Deep Learning Framework for Predicting Football Match Outcomes and Evaluating Team Performance

This study presents a novel predictive framework for estimating football match outcomes and assessing team performance through a new metric called the Success Score. This metric integrates Expected Goals (xG) with actual scoring results to provide a comprehensive evaluation of both opportunity creation and offensive execution. A Deep Neural Network (DNN) was trained on three seasons (2020–2021, 2021–2022, 2022–2023) of match data from four major European leagues, incorporating features such as tactical play styles, rolling performance averages, and team quality indicators. The model demonstrated strong predictive performance, achieving a Mean Absolute Error (MAE) of 0.3142 and an R² score of 0.8592 in cross-validation. It also generalized effectively to out-of-sample matches from the 2024–2025 season. Furthermore, the Success Score allowed for the classification of outcomes, enabled outcome prediction with 73.30% accuracy for Win vs. Not Win and 75.13% for Lose vs. Not Lose. Beyond predictive accuracy, the framework offers interpretable insights into team dynamics, revealing patterns of overperformance and underperformance. Case studies of FC Barcelona and Manchester United illustrate their practical utility for tactical analysis and strategic planning. This scalable, data-driven approach advances modern football analytics by supporting continuous performance monitoring and informed decision-making.