HYBRID MACHINE LEARNING AND FEDERATED LEARNING FRAMEWORK WITH EXPLAINABLE AI FOR PERSONALIZED STROKE TELEREHABILITATION IN RESOURCE-CONSTRAINED SETTINGS

Abstract

Background: Stroke rehabilitation in resource-constrained environments is limited by restricted access to specialized care, fragmented healthcare systems, and concerns related to data privacy and infrastructure. Telerehabilitation offers a scalable solution; however, existing systems lack predictive intelligence, personalization, and interpretability required for clinical adoption.Objective: This study aims to develop and evaluate a hybrid machine learning and federated learning framework for stroke telerehabilitation that ensures high predictive performance, data privacy, and model interpretability. Methods: A hybrid machine learning approach was implemented using Support Vector Machine, Decision Tree, Random Forest, and Artificial Neural Network models. A proposed hybrid ensemble model was developed using a weighted soft voting strategy to improve predictive accuracy. Federated learning was integrated using the Federated Averaging (FedAvg) algorithm across multiple simulated nodes to enable privacy-preserving distributed training. Data preprocessing included normalization and imputation, and model performance was evaluated using accuracy, precision, recall, F1 score, and ROC-AUC. Robustness was assessed using 10-fold cross-validation, with results reported as mean ± standard deviation. Results: The Random Forest model achieved the highest performance among individual models (accuracy: 89.3% ± 1.2). The proposed hybrid ensemble model demonstrated superior performance with an accuracy of 92.1% ± 1.0 and ROC-AUC of 0.93 ± 0.01. The federated hybrid ensemble model achieved comparable performance (accuracy: 90.8% ± 1.3; ROC-AUC: 0.92 ± 0.01), indicating minimal performance loss under distributed conditions. Cross-validation results confirmed model stability, while explainable AI techniques identified motor function score and therapy adherence as key predictors. Conclusion: The integration of hybrid machine learning, federated learning, and explainable AI provides an effective and scalable solution for stroke telerehabilitation. The proposed framework achieves high predictive performance while ensuring data privacy and interpretability, making it suitable for deployment in resource-constrained healthcare environments. Further validation using real-world clinical data is required to confirm its practical applicability.

Keywords : Stroke Telerehabilitation; Hybrid Ensemble Learning; Federated Learning; Explainable AI (XAI); SHAP; LIME; Privacy-Preserving Machine Learning; Stroke Rehabilitation Prediction; Resource-Constrained Healthcare.