Evaluating the Predictive Performance of AI in Football Match Forecasting: A Statistical and Comparative Analysis Across European Leagues

Abstract

This study evaluates the predictive accuracy of an AI-based model in forecasting football match outcomes and in-game statistics, specifically ball possession and pass accuracy. A total of 200 matches from top-tier leagues in ten European countries and two international club competitions were analyzed. Predictions were collected on match-day mornings using real-time web searches and compared with actual results. The AI model correctly predicted 55.5% of match outcomes, including 25 exact scorelines. Prediction accuracy varied by league, with the highest rates in Italy (70.6%) and the UEFA Champions League (66.7%), and the lowest in England (16.7%) and Portugal (17.6%). A Chi-Square test indicated a statistically significant association between AI predictions and actual results (χ² = 46.520, df = 4, p < 0.001), suggesting predictions were not random but reflected underlying patterns. Pearson correlation analysis revealed moderate relationships between predicted and actual in-game statistics, particularly for pass accuracy (r = 0.626 for away teams) and ball possession (r = 0.591 for away, r = 0.581 for home teams). Findings indicate that while AI can offer valuable insights, its reliability is inconsistent across leagues and metrics. AI models tend to perform better in structured, data-rich contexts, while unpredictable leagues present greater forecasting challenges. Future research should integrate real-time match data, advanced machine learning techniques, and sentiment analysis from social media and expert commentary to enhance predictive performance and bridge the gap between computational models and real-world football dynamics.