Tennis match outcome prediction using machine learning and in-match statistics

Machine learning has become a cornerstone of sports analytics, including tennis, which is an ideal arena for its application due to the abundance of comprehensive tennis data. This study investigates machine learning algorithms in forecasting tennis match results, comparing their performance against traditional methods. Using a dataset spanning from 2000 to 2022, encompassing a wide range of player statistics and match details, we employed multiple prediction models, Logistic Regression, Random Forests, TabNet, and XGBoost. We explored the role of in-match statistics as key predictors, unveiling their significance in determining match outcomes. Rather than deploy feature selection and feature engineering techniques, we relied on machine tree-based learning models that can handle high-dimensional data while mitigating the risk of overfitting. We compared our results against various baselines, including the Glicko model, and demonstrate a superior performance. This highlights the potential of machine learning techniques, particularly XGBoost, in elevating predictive accuracy in tennis match forecasting. Our study contributes to the growing body of research in sports analytics by emphasizing the importance of in-game performance data in predicting tennis match outcomes. By integrating machine learning approaches, we demonstrate the potential for more accurate and insightful match predictions, benefiting both fans and players alike.