Filter-Based Feature Selection Using Machine Learning Algorithms and Physical Indicators for CAD Prediction

Abstract

Coronary artery disease (CAD) is a significant threat, especially in developing nations, demanding timely detection for effective intervention. Traditional tools like electrocardiography and angiography have been pivotal for CAD prediction, but the rise of machine learning (ML) introduces new possibilities for computer-aided diagnosis. This chapter unveils an innovative CAD detection method that synergizes filter-based feature selection with ML classifiers, notably enhancing their efficacy. Various filter-based techniques generate distinct feature subsets, and features commonly selected through majority voting significantly improve the accuracy of the k-nearest neighbor (KNN) classifier, rising from 88.31% to 89.61%. This underscores the positive impact of feature selection on the KNN classifier's accuracy, promising utility in broader medical datasets. The proposed method holds potential for advancing CAD diagnosis and prognosis, particularly in resource-constrained settings, by minimizing false positives.