Applying Machine Learning for Analyzing Shooting Importance in Modern Pentathlon

This study aimed to analyze and quantify the relative importance of each shooting series (Series 1 through 4) to variations in total shooting performance during the laser run event in Modern Pentathlon using machine learning (ML) models and interpretability methods. Individual shooting times (n = 1453) were collected from six international competitions hosted by UIPM in 2024, and 2-ML models, Random Forest and XGBoost, were trained to predict total shooting time. To interpret model results, two interpretability methods—Permutation Importance and SHAP (SHapley Additive exPlanations)—were applied to the trained models, and their performance was evaluated using Mean Absolute Error (MAE) and R-squared (R2) with 5-fold cross-validation. Series 4 shooting consistently exhibited the highest importance in explaining variations in total shooting time for both male and female athletes, with the XGBoost model interpreted using SHAP achieving the highest predictive accuracy (R2 = 0.97). Additionally, Permutation Importance identified Series 1 as particularly influential among male athletes, addressing potential biases in Random Forest’s Mean Decrease Impurity measure. These findings highlight the importance of managing concentration, rapid recovery, and psychological pressure during the final shooting stage, alongside the significance of stability in earlier series, providing actionable, data-driven strategies for training and psychological preparation in Modern Pentathlon athletes.