A review of metaheuristic gravity search algorithms and hybrid methods based on neural networks for the diagnosis of coronary heart disease (improvement, comparison and performance analysis)

Cardiovascular disease remains a leading global health challenge, necessitating advanced diagnostic tools for early detection. This study introduces novel hybrid models combining gravitational search (GSA) and particle swarm optimization (PSO) with multilayer perceptrons (MLP) to significantly improve cardiac diagnosis accuracy. We propose two architectures: GSA-MLP and PSO-MLP, where evolutionary algorithms optimize MLP weights before fine-tuning. Evaluated on the Cleveland Clinic dataset through ۵-fold cross-validation, our hybrid models demonstrated remarkable performance gains. The GSA-MLP achieved ۹۱.۵۲% accuracy (۸۹.۲۳% recall), reducing critical false negatives by ۷۱% compared to standard MLP (۸۰.۸۰% accuracy). Similarly, PSO-MLP attained ۸۹.۶۳% accuracy with ۸۶.۷۵% recall. Confusion matrix analysis revealed GSA-MLP correctly identified ۲۶.۴ true positives among ۵۹ test cases while maintaining only ۱.۶ false negatives – a clinically crucial improvement minimizing missed diagnoses. These results establish that hybridization enables superior navigation of complex medical decision boundaries: GSA's mass-interaction mechanism outperformed PSO's velocity-based approach by ۱.۸۹% accuracy, while both hybrids surpassed standalone algorithms by ۷-۹%. The ۱۱% accuracy enhancement over conventional neural networks demonstrates our framework's ability to overcome local optima limitations through strategic weight initialization. This research provides evidence that evolutionary-neural integration represents a paradigm shift for high-stakes medical diagnostics, with potential applications extending to other complex classification domains where precision impacts outcomes.

کلیدواژه ها:

Diagnosis of Coronary Artery Disease (CAD) ، Gravitational Search Algorithm (GSA) ، Particle Swarm Optimization (PSO) ، Multilayer Perceptron Neural Network (MLP) ، Hybrid PSO-CNN Model ، Diagnostic Accuracy ، Dimensionality Reduction ، Non-invasive