INTEGRATING FEATURE OPTIMIZATION WITH MACHINE LEARNING FOR HEART DISEASE DETECTION

Abstract

Cardiovascular diseases (CVDs) contribute significantly to global morbidity and mortality in the world, and need proper and timely diagnostic assistance. Machine learning (ML) techniques have demonstrated high capabilities in predicting heart disease but their accuracy is highly reliant on the relevance and quality of features. This paper provides a critical analysis of three bio-inspirational optimizer, Particle Swarm Optimization (PSO), Grey Wolf Optimizer (GWO), and Firefly Algorithm (FA) in the context of the Jupyter Heart Disease (JUH) sample. The five ML classifiers which were analyzed include SVM, random forest, decision tree, naive bayes and KNN with and without optimization. Moreover, a hybrid PGF-Optimizer that combines the global exploration of PSO with the convergence of GWO, using a leader, and the local refinement of FA is suggested. As demonstrated in the experiment, the optimization of features has been proven to be able to continuously enhance the performance of the classifier and the Hybrid PGF model has not only the best accuracy and the best recall, but also the best ROC-AUC among all models. SHAP-based explainability is useful in improving interpretability as it can provide significant clinical features that are used to make predictions. The presented results show that hybrid optimization can enhance the cardiac risk prediction using machine learning and offers a promising paradigm of clinical decision-support systems in practice.