XAI-Based Ensemble Model for Human Body Performance Classification

This research presents a comprehensive machinelearning approach for predicting human body performance classes using multiple classification models and ensemble techniques. The study evaluates six models: Random Forest, Support Vector Machine (SVM), XGBoost, Gradient Boosting (optimized), TabNet, and various ensemble methods—specifically Voting and Stacking. Extensive hyperparameter tuning was conducted to enhance the performance of these models. Among the individual models, XGBoost achieved the highest accuracy of 0.766. However, the optimized Stacking model, which combines XGBoost, SVM, and Random Forest with a neural network as the metalearner, outperformed all other models, reaching an accuracy of 0.776 and an F1-score of 0.780. Additionally, explainable AI tools such as SHAP and LIME were utilized to interpret predictions made by the models across all classes, providing transparency and insights into feature importance. The results demonstrate that ensemble techniques, mainly Stacking, significantly enhance classification performance in predicting human body performance.