Feature Importance Analysis: Predicting Tennis Match Outcomes

In today's data-driven world, tennis offers extensive player and match statistics that can be utilized for match outcome predictions, benefiting both betting markets and training programs. However, studies differ on whether player rank or serve strength alone suffices for accurate predictions. This study evaluates XGBoost and Random Forest models trained on match and player statistics, rank alone, serve-related features alone, and rank with service-related features only. Results show that both models achieved at least 80% accuracy using all features or serve-related features, compared to around 60-65% when using rank alone. Adding player rank to serve-related features did not significantly improve model performance. These findings highlight serve strength as the most critical predictor of outcomes in men’s singles tennis matches on the ATP circuit.