CardioMetaHybridOptimizer as a behaviorally adaptive multi-phase metaheuristic framework for interpretable...

Area of Science:

• Cardiology and Computational Health Science

Background:

• Cardiovascular disease (CVD) prediction is challenged by complex, high-dimensional clinical data.
• Effective feature selection is crucial for accurate risk assessment and clinical decision support.

Purpose of the Study:

• To introduce the Cardio Meta Hybrid Optimizer (CMHO) framework for enhanced feature selection and predictive accuracy in cardiac risk assessment.
• To develop a robust computational tool for navigating high-dimensional data in cardiology.

Main Methods:

• The CMHO framework integrates Lion Optimization (LO), Marine Predators Algorithm (MPA), and Manta Ray Foraging Optimization (MRFO) with adaptive switching, dynamic mutation, and iterative local search (ILS).
• A CNN-LSTM architecture was employed for classification, validated using stratified tenfold cross-validation on five benchmark datasets.
• Performance was statistically compared against established methods like RFE, GA, PSO, GWO, and Lasso using ANOVA.

Main Results:

• The CMHO-integrated CNN-LSTM model achieved 96.1% accuracy, outperforming traditional methods by 3%-5% (p < 0.05).
• The framework demonstrated high stability (Stability Selection Index > 0.90) and clinical interpretability, identifying key biomarkers like thalassemia, chest pain type, and maximum heart rate.

Conclusions:

• The CMHO framework offers a robust, interpretable, and stable computational approach for cardiovascular risk assessment.
• This tool enhances clinical decision support by effectively managing high-dimensional data and identifying relevant biomarkers across diverse patient populations.