Application of computer simulation technology in the optimization and performance evaluation of smart sports teaching systems in universities

This study proposes a computer-simulation-driven framework for optimizing and evaluating smart sports teaching systems in universities. Using MHEALTH as the main multimodal wearable-sensor dataset, and UCI HAR Smartphone and Daily and Sports Activities as external validation datasets, the framework integrates wearable-sensor preprocessing, window-level feature construction, activity recognition, teaching-state indicator construction, sensor-ablation analysis, scenario simulation and optimization, ...statistical validation, and computational implementation-feasibility assessment. Activity Load Index, movement stability, and a sensor-derived safety proxy were constructed to translate acceleration, gyroscope, magnetometer, and ECG signals into interpretable teaching-state representations. To address subject-dependence and potential overestimation from a fixed train/test split, classification performance was re-evaluated using leave-one-subject-out validation. In the realistic MHEALTH setting including the dominant null class, Gaussian Naive Bayes achieved the highest mean Macro F1 of 0.652, while tuned Extra Trees achieved an accuracy of 0.819 and a Macro F1 of 0.563. In the activity-only setting, Random Forest and Extra Trees achieved mean Macro F1 scores of approximately 0.95, indicating strong but not universally perfect subject-independent recognition. External validation produced Extra Trees accuracies of 0.941 on UCI HAR and 0.964 on Daily Sports. Sensitivity analyses showed that the principal high-load activity assignments remained stable under plausible variations in Activity Load Index weights and thresholds. Formal constrained optimization identified a 25% low, 65% moderate, 8% high, and 2% very-high state allocation as the utility-maximizing solution, corresponding to a 13.7% simulated utility gain over the observed baseline. These results support the use of the proposed framework as an interpretable, subject-independent decision-support approach for evaluating wearable-sensor-based sports teaching scenarios.