Interpretable Skill Prioritization for Volleyball Education via Team-Stat Modeling

Abstract

Allocating scarce training time among technical elements is a constant challenge for volleyball coaches. This study operationalizes that decision as an interpretable, data-driven ranking problem: given season-level team statistics, predict team success and quantify which skills contribute most. We constructed a Division I women’s volleyball team-season dataset for 2022–2023 by aggregating publicly available NCAA match statistics into per-set rates and efficiency measures (e.g., hitting percentage, opponent hitting percentage faced). The final dataset contains 344 teams with 8 skill-related predictors and the target win–loss percentage. We evaluated multiple regression models on repeated random splits. A Gradient Boosting Regressor achieved the best generalization (RMSE 0.068 ± 0.006; R² 0.887 ± 0.026). To translate prediction into coaching action, we computed global skill importance using permutation importance and SHAP values. Both methods showed high ranking stability and identical top-three priorities: suppressing opponent hitting percentage (defense), own hitting percentage (attack efficiency), and kills per set (terminal finishing). Based on these effects, we propose a teaching priority list emphasizing (1) reducing opponent efficiency via coordinated serve–block–defense, (2) raising attack efficiency, and (3) building terminal attacking capacity. The framework is reproducible, model-agnostic, and directly maps team statistics to curriculum design.