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

Area of Science:

• Artificial Intelligence in Healthcare
• Machine Learning for Predictive Analytics
• Computational Health Informatics

Background:

• Smart health systems leverage machine learning (ML) and electronic health records (EHRs) for disease prediction and personalized care.
• Existing ML models require optimization for enhanced accuracy in processing large medical datasets.
• Effective forecasting is crucial for timely interventions and improved patient management.

Purpose of the Study:

• To propose and evaluate novel hybrid machine learning models for intelligent healthcare.
• To optimize Stacking Classifier (StackingC) and Bagging Classifier (BaggingC) using prairie dog optimization (PDO) and sooty tern optimization algorithm (STOA).
• To assess the predictive accuracy and generalization capabilities of the optimized models.

Main Methods:

• Development of hybrid models combining StackingC and BaggingC with PDO and STOA optimization.
• Attribute selection and model accuracy maximization using optimization algorithms.
• Sensitivity analysis of the best hybrid models (STPD and STST) for risk prediction consistency.

Main Results:

• StackingC demonstrated superior overall accuracy (0.979) compared to BaggingC (0.958).
• Optimized hybrid models STPD and STST achieved high overall accuracy (0.985 and 0.980, respectively).
• StackingC showed better generalization performance on test data (0.942) despite BaggingC's higher training accuracy (0.961).

Conclusions:

• Hybrid ML approaches, optimized with PDO and STOA, significantly enhance predictive analytics in smart healthcare.
• The proposed models offer accurate and robust disease prediction, contributing to improved patient outcomes.
• Sensitivity analysis confirms the consistency and reliability of these hybrid models for risk prediction.