Predicting soccer matches with complex networks and machine learning.

This article investigates the use of complex network analysis of passing patterns combined with traditional match statistics to predict soccer match outcomes. By constructing player passing networks and extracting network metrics, the study applies machine learning models—including Logistic Regression, Random Forest, and XGBoost—to predict wins and losses across multiple major leagues and international tournaments. Results indicate that models based solely on passing networks perform comparably to those using conventional statistics, while a combined model integrating both data types achieves improved predictive accuracy, with the Random Forest model reaching approximately 71.5% accuracy and an AUC of 0.77. Additionally, the study finds no significant differences in playing styles among the analyzed European leagues, suggesting that predictive models trained on one league can generalize to others, though the English Premier League showed higher predictability. The research highlights the potential of integrating network science and sports analytics for enhanced understanding and forecasting of soccer match outcomes.