Abstract
Abstract
Objective evaluation of single-shot execution in air pistol shooting is limited by the lack of interpretable methods that reproduce coaches’ technical judgments. We developed a random forest classifier using SCATT aiming trajectories and virtual shot locations from 2,857 shots by two elite athletes, labeled as good or non-good executions by professional coaches. Candidate features were organized into coach-informed groups, screened at group and variable levels, and interpreted using Shapley Additive Explanations (SHAP). External consistency was assessed in 1,566 shots from two additional athletes without retraining or threshold adjustment. On the internal test set, the final model achieved balanced accuracy of 0.8450, good-execution F1 of 0.7532, and area under the receiver operating characteristic curve (ROC-AUC) of 0.9175. The corresponding values in the additional athletes were 0.8187, 0.7513, and 0.9349. The most influential features described the relationship of the virtual impact point and terminal blue trajectory to the yellow stability region, together with the green-to-yellow transition. These findings indicate that SCATT trajectory features can reproduce coach-defined execution-quality labels and provide technically meaningful explanations for post-shot review. Because only four athletes were included, broader validation is required.