[Paper Review] Predicting Tennis Serve directions with Machine Learning

Ying Zhu, Ruthuparna Naikar|arXiv (Cornell University)|Feb 26, 2026

Sports Analytics and Performance0 citations

TL;DR

The authors build ML models to predict professional players’ first-serve directions (three directions per side) using Match Charting Project data, achieving about 49% accuracy for male players and 44% for female players, based on contextual features.

ABSTRACT

Serves, especially first serves, are very important in professional tennis. Servers choose their serve directions strategically to maximize their winning chances while trying to be unpredictable. On the other hand, returners try to predict serve directions to make good returns. The mind game between servers and returners is an important part of decision-making in professional tennis matches. To help understand the players' serve decisions, we have developed a machine learning method for predicting professional tennis players' first serve directions. Through feature engineering, our method achieves an average prediction accuracy of around 49\% for male players and 44\% for female players. Our analysis provides some evidence that top professional players use a mixed-strategy model in serving decisions and that fatigue might be a factor in choosing serve directions. Our analysis also suggests that contextual information is perhaps more important for returners' anticipatory reactions than previously thought.

Motivation & Objective

• Understand factors influencing professional players' first-serve direction choices.
• Evaluate prediction accuracy of machine learning models on serve direction outcomes.
• Investigate whether contextual information, fatigue, and anxiety explain serve-direction decisions.
• Explore feature importance to shed light on decision-making in serving.

Proposed method

• Use ground-truth serve direction data from the Match Charting Project as labels.
• Engineer features from point-by-point data including prior serve counts, serve percentages, fatigue measures, and performance anxiety indices.
• Train multiple models (Multinomial Logistic Regression, Decision Tree, Random Forest, SVM, Neural Network) per player with 70/30 train-test split.
• Evaluate performance separately for deuce and ad sides and for male and female players.
• Identify feature importance via Decision Tree analysis to interpret servedirectional decisions.

Experimental results

Research questions

• RQ1 Can machine learning predict individual players’ first-serve directions with contextual features alone?
• RQ2 What is the predictive accuracy across players and gender when using three serve directions per side?
• RQ3 Which features most influence serve-direction choices (e.g., prior counts, fatigue, anxiety)?
• RQ4 Do results support mixed-strategy behavior and the role of fatigue in serve decisions?

Key findings

• Average prediction accuracy is about 49% for male players on the deuce side and about 50% on the ad side.
• Average prediction accuracy is about 44% for female players on the deuce side and about 45% on the ad side.
• Prediction accuracy is consistent across multiple ML methods for a given player, suggesting robustness of results.
• Feature importance analysis highlights cumulative counts of first serves to each direction, run index before the current point, and first-serve percentage as influential.
• Results provide evidence that top players may employ mixed strategies and that fatigue may affect serve-direction choices.
• Contextual information alone can yield reasonable predictions for returners’ anticipatory reactions.