A hybrid machine learning-based recommendation system for personalized prevention and management of chronic diseases

Abstract

Background: Chronic diseases, including diabetes, cardiovascular diseases, hypertension, and chronic obstructive pulmonary disease (COPD), represent a major global health burden, accounting for approximately 74% of deaths worldwide. Conventional disease management guidelines often adopt generalized approaches that inadequately address patient heterogeneity, lifestyle factors, and comorbidities. Therefore, intelligent, evidence-based systems capable of delivering personalized preventive and treatment recommendations are urgently needed. Objective: This study proposes a hybrid machine learning-based recommendation system integrating collaborative filtering, content-based filtering, Random Forest, and XGBoost algorithms to generate personalized chronic disease prevention and management plans. Methods: The framework utilizes five publicly available medical datasets comprising 6,682 patient records covering multiple chronic diseases. A weighted ensemble fusion method integrates outputs from recommendation and classification modules. System performance was evaluated using stratified 10-fold cross-validation based on accuracy, precision, recall, and F1-score. Results: The proposed hybrid model achieved an overall accuracy of 94.3%, precision of 93.1%, recall of 93.7%, and F1-score of 95.2%, outperforming the baseline models. XGBoost and collaborative filtering achieved accuracies of 89.1% and 81.2%, respectively, demonstrating the effectiveness of the hybrid fusion strategy. Conclusion: The proposed architecture improves personalization and predictive performance compared with standalone approaches and demonstrates strong potential for integration into clinical decision-support systems and digital health applications, supporting proactive and patient-centered chronic disease management.