Accelerating the performance of machine learning classifiers using bacterial colony optimization for heart disease...

Area of Science:

• Biomedical Engineering
• Computational Biology
• Artificial Intelligence in Medicine

Background:

• Cardiovascular disease (CVD) is a leading cause of global mortality, necessitating improved diagnostic methods.
• Conventional CVD diagnosis faces challenges like misdiagnosis, delayed treatment, and high costs.
• Machine learning (ML) offers potential for early CVD detection but struggles with parameter tuning and local minima.

Purpose of the Study:

• To develop a bio-inspired optimization framework to enhance the accuracy and efficiency of ML-based CVD diagnosis.
• To address computational challenges in ML model parameter tuning and local minima susceptibility.
• To improve early cardiovascular risk detection through a robust decision-support tool.

Main Methods:

• Employed Bacterial Colony Optimization (BCO) to tune hyperparameters of ten ML classifiers.
• Integrated Principal Component Analysis (PCA) for feature dimensionality reduction and multicollinearity.
• Validated models on Cleveland Heart Disease (CLE) and IEEE DataPort (HGR) datasets using 5-fold Cross-Validation.

Main Results:

• The proposed framework significantly improved prediction performance over baseline models.
• Bacterial Colony Optimization-optimized Random Forest achieved 92.02% accuracy on the HGR dataset.
• Bacterial Colony Optimization-optimized Support Vector Machine reached 85.79% accuracy on the CLE dataset, with enhanced stability.

Conclusions:

• The framework effectively enhances cardiovascular disease diagnosis accuracy and stability.
• Bio-inspired optimization mitigates local minima issues, offering a reliable tool for early risk detection.
• The approach provides a statistically robust and clinically valuable decision-support system for cardiovascular health.