Enhancing predictive accuracy in smart health systems through hybrid machine learning models and sensitivity analysis

Abstract

Smart health systems integrate advanced machine learning (ML) approaches to strengthen disease prediction, patient monitoring, and personalized healthcare solutions. They are founded on electronic health records (EHRs), mobile health (M-Health), and electronic medicine (E-Medicine) to process vast amounts of medical data in a timely and effective way. In this study, a new forecasting model is proposed employing Stacking Classifier (StackingC) and Bagging Classifier (BaggingC) models, which were optimized employing prairie dog optimization (PDO) and the sooty tern optimization algorithm (STOA). These optimization methods maximize attribute selection and model accuracy, ensuring strong and accurate forecasts. It is apparent from the results that StackingC is superior to BaggingC in overall accuracy at 0.979 against BaggingC at 0.958. Although BaggingC performed better at training accuracy (0.961), StackingC performed better at overall generalization when tested (0.942). We further introduce a sensitivity analysis of the best hybrid models, STPD and STST, to demonstrate their consistency in risk prediction. STPD performed the best at overall accuracy at 0.985, followed by STST at 0.980. These findings testify to the excellence of hybrid ML approaches in intelligent healthcare situations to ensure improved patient outcomes through accurate and robust prediction. This study contributes to predictive analytics in health care by optimizing model and sensitivity analysis approaches.

Ähnliche Arbeiten