Coronary artery disease classification using support vector machines tuned via randomized search cross-validation

Abstract

Coronary artery disease outstands health problem that causes high mortality in the world population. This disease brings with it fateful problems such as heart attack and heart failure in patients with cardiovascular problems. Early diagnosis of coronary artery disease is essential for the timely administration of the right treatment and reduction of mortality. Angiography is the most preferred method for CAD detection. However, the complications and costs of this method have led researchers to forage alternative methods through machine learning algorithms. By developing a machine learning model with high generalization ability, prediction errors can be minimized. Thus, these models could potentially be useful for specialist physicians in the effective detection of coronary artery disease. The main focus of this study is to perform coronary artery disease detection with improved support vector machines. k-fold cross-validation experiments were performed on the Z-Alizadeh Sani dataset to evaluate the performance of the models. According to the results obtained, support vector machines with randomized search cross-validation provided the best performance when compared to other models. 87.102% average accuracy, 91.176% average sensitivity, 90.852% average precision, 76.996% average specificity, and also 8.824% average false negative rate obtained by 5-fold cross-validation competes with the known approaches in the literature.