A Framework for Forecasting Professional Tennis Matches Using ATP Statistics and Betting Odds

We propose a machine learning framework to forecast professional tennis outcomes using a 57,000-match dataset (2003–2024). By combining engineered ATP statistics with pre-match betting odds, the model captures both player dynamics and market expectations. Comparative results show that ATP-based features alone achieve performance within 1-2% of odds-only models, confirming the power of domain-driven engineering. Our final XGBoost model reaches nearly 70% accuracy on unseen matches using only pre-match data.