Accurate Recognition of Coronary Artery Disease by Applying Machine Learning Classifiers

Being one of the most common causes of death, coronary artery disease (CAD) has been an area of interest for many researchers throughout the decades. Recognition of CAD disease in the early stage can save a significant amount of lives. To achieve this goal, previously, many researchers have contributed significantly to the classification of CAD disease. Many of the studies achieved higher classification accuracy with a different number of features. However, in all the cases, either the number of selected features was too high or the accuracy was too low. In this study, we considered an extensive preprocessing approach followed by five feature selection algorithms (Boruta, lasso, minimum redundancy maximum relevance, recursive feature elimination based on random forest and chi-square test). We also considered the common features selected by all of the techniques. After that, seven classifiers (Naïve Bayes, decision tree, k-nearest neighbors, logistic regression, random forest, support vector machine, and neural network) were applied and the performances were measured in terms of accuracy, sensitivity, specificity, f1-score, MCC, and AUC values. With an overall accuracy of 92.31% and considering only 21 features, we showed that our combination of accuracy and selected features produced a better approach to classify CAD disease.