Abstract High-fidelity simulation is vital to competency-based medical and nursing education, yet real-time telemetry for predictive feedback is rare. Current analytics remain predominantly descriptive, hindering the timely detection of learner challenges and constraining opportunities for adaptive instruction. This study introduces a real-time machine learning framework to predict simulation performance and provide transparent, actionable feedback. Five supervised models were implemented in Python using the JIGSAWS Needle-Passing task: XGBoost, Temporal Convolutional Networks (TCN), Random Forest, Artificial Neural Networks, and L1-Regularized Logistic Regression. Input features included tool trajectories, gesture segmentation, timing metrics, and motion smoothness. XGBoost achieved the highest performance (F1 = 0.78, AUC = 0.87), followed by TCN (F1 = 0.71), with superior temporal modeling intermediate learners. SHAP-based explanations identified tool velocity, error rate, and motion smoothness as the most predictive features. Ablation studies confirmed behavioral features as the primary contributors to model accuracy. Regression results further supported XGBoost’s reliability (R² = 0.81). The proposed framework enables sub-second inference, promotes early identification of at-risk learners, and delivers pedagogically meaningful insights. Integrating explainable AI into simulation environments transforms them from static assessment tools into dynamic, adaptive learning systems. The framework offers a scalable, cross-disciplinary path toward precision education, addressing global calls for equitable, data-driven, and personalized healthcare education. Future research should prioritize scaling the framework to diverse learner populations, embedding it within real-time simulation workflows, incorporating multimodal data streams, and evaluating its transferability across clinical domains to support intelligent, adaptive feedback in medical education.
Bulus Bali (Tue,) studied this question.