Quantifying the tactical efficiency of shot redirection in elite men’s tennis rallies: a stroke-level machine learning framework

Shot redirection is a common but technically demanding tactical action in elite tennis rallies. Its immediate tactical value is difficult to quantify using existing performance indicators. This study developed a stroke-level machine-learning framework to evaluate the tactical efficiency of shot redirection in elite men’s baseline rallies. Thirty Grand Slam men’s singles matches from 2023 to 2025 were analysed. The dataset contained 33,975 strokes from baseline rallies on hard, clay, and grass courts. Each stroke was labelled as Attack, Neutral, or Defense. An XGBoost model estimated expected rally status (xRS) as the predicted probabilities of the three states. The probabilities were converted into expected stroke performance value (xSPV). Redirection efficiency was represented by the difference in mean xSPV between redirected and direction-maintained strokes across surface type, stroke technique, Shot Location Depth and movement demand. Criterion-related validity was assessed against point outcome. Robustness was examined across 21 alternative utility-weight specifications. Point-winner versus point-loser comparisons were corrected using the Benjamini–Hochberg false discovery rate procedure. The xRS model achieved an Accuracy of 0.7811, a Weighted F1-score of 0.7802, and a Log-loss of 0.4046 on the held-out test set. SHAP analysis identified position–technique interaction, movement demand, shot direction, stroke technique, stroke position, shot location depth, and opponent-related context as the main contributors to rally status. Redirected shots increased the mean predicted probabilities of Attack and Neutral and reduced the probability of Defense. ΔxSPV analyses showed that redirection was not uniformly beneficial. Positive effects were mainly observed for forehand and backhand strokes, under low-to-moderate movement demands, and when sufficient shot depth was maintained. Higher xSPV was associated with greater odds of winning the point (OR = 1.079 per 0.1-unit increase), and the association was strongest near point termination. The main ΔxSPV patterns remained stable across 21 alternative utility-weight specifications. The tactical efficiency of shot redirection depends on the interaction between directional change, stroke execution, movement demand, and landing-depth quality. The xRS–xSPV framework provides a context-sensitive measure of immediate tactical value. Its association with point outcome and stability across alternative utility weights support its use for retrospective tactical evaluation in elite tennis.