The quality of traditional Chinese medicine (TCM) training directly affects the formation of syndrome differentiation thinking, the mastery of operational skills and the improvement of subsequent clinical competence of students. However, the existing evaluation methods mostly rely on teachers ‘experience scoring and outcome assessment, which has problems such as insufficient utilization of indicators, insufficient integration of process information and unstable quality stratification. Therefore, this paper constructs the quality evaluation decision model of traditional Chinese medicine training based on improved Random Forest, and integrates multi-dimensional indicators such as theoretical test scores, manipulation standard degree, operation proficiency, case analysis integrity, classroom participation and teaching feedback into the unified feature space. Through feature screening, category weight adjustment and parameter optimization, the recognition ability of the model for complex teaching data was improved. The improved Random Forest was further compared with Decision Tree, standard Random Forest and XGBoost. The results show that the Accuracy, Precision, Recall and F1 values of the improved Random Forest reach 0.923, 0.911, 0.896 and 0.903, respectively, and the verification error is reduced to 0.112 at the 80th iteration. The overall performance is better than that of the control model. The empirical analysis found that the accuracy of syndrome differentiation judgment, the standard of manipulation, the proficiency of operation and the integrity of case analysis were the core factors affecting the quality stratification of TCM training. The research shows that the improved Random Forest can effectively reveal the key variables in the formation of traditional Chinese medicine training quality, and provide a new technical path for the optimization of training teaching, hierarchical guidance and digital teaching management.
