Machine Learning-Based Classification of Alertness Levels in Elite Shooting Athletes Using Heart Rate Variability.

This study aims to develop a predictive model for alertness levels in elite shooting athletes by analyzing heart rate variability (HRV) dynamics under simulated competitive stress. 83 national-level shooting athletes completed a 60-minute Psychomotor Vigilance Task (PVT) protocol designed to mimic the sustained attentional demands of a competition, while HRV data were continuously recorded. Pearson correlation analysis identified HRV features associated with behavioral performance. Key predictors were selected via recursive feature elimination with Random Forest. Four machine learning algorithms-Support Vector Machine (SVM), Random Forest (RF), XGBoost, and AdaBoost-were employed to construct classification models for alertness. Model performance was evaluated using accuracy, precision, recall, F1-score, and the area under the ROC curve (AUC). SHAP analysis was applied to interpret feature contributions. The binary classification framework (optimal vs. sub-optimal alertness) demonstrated superior reliability over multi-class approaches. The AdaBoost model achieved the best performance, with an accuracy of 0.75, an F1-score of 0.73, and an AUC of 0.77. SHAP analysis revealed that the very low frequency percentage (VLF%) was the most critical predictor, followed by the SD2/SD1 ratio. Notably, elevated VLF% values were associated with lower alertness levels. The binary classification model, integrating key HRV indices (notably VLF%) with the AdaBoost algorithm, can effectively distinguish alertness levels in shooting athletes during simulated competitive stress. This approach provides a validated, non-invasive tool for objective psychophysiological monitoring in training, offering actionable insights for pre-competition readiness assessment.