Simplified Deep Neural Network Models for Cardiovascular Disease Classification

Abstract

Cardiovascular diseases encompass a range of conditions affecting the heart and blood vessels. Given their global impact, early detection of these diseases is crucial for saving lives and effectively managing morbidity and mortality. One such effective approach is to leverage deep learning to enhance classification performance in heart disease prediction (HDP). This paper presents two new simplified deep neural network (SDNN) models to assist cardiologists and vascular doctors in diagnosing heart disease that achieve 100% accuracy with feature selection, outperforming complex hybrid models of related works. The proposed models are tested on combined Kaggle datasets containing consistent reports of 918 people, including heart disease and non-heart disease reports. The models are investigated with/without applying feature selection methods and compared with different machine learning classifiers and recently proposed SDNN models. The experimental results show how the proposed SDNN models outperform other ML-based and DL-based classifiers in terms of accuracy and structure�s complexity. The proposed model, SDNN-HDP1, with dropout layers achieves 94.086% accuracy, while the second proposed model, SDNN-HDP2, without dropout layers achieves 95.69% accuracy without feature selection. The results of SDNN-HDP1 and SDNN-HDP2 reach 100% in terms of accuracy, precision, and recall with feature selection.