Predicting Heart Disease with Multiple Classifiers

Abstract

Heart disease accounts for several deaths worldwide and is the mainspring of cardiovascular morbidity. It is increasingly ravaging human lives—about a hundred million in the previous decade. Given that data is generated daily by several sources and in various formats, its timely, accurate, and cost-efficient prediction is imperative for clinical reforms. Machine learning is reputable for its effectiveness in the prediction of medical conditions by use of medical data. The complex correlations present in medical datasets increase the complexity of such predictions. A novel hybrid technique is proposed to improve the prediction of heart disease in this document. This technique focuses on the reduction of false negatives for the betterment of patient care. The proposed technique assigns weights to four classifiers each built with one of the four reputable algorithms—decision tree, random forest, K-nearest neighbor, and logistic regression. The final class of a new instance is that predicted by the maximum weighted sum of predictions from the classifiers. This technique is compared with already existing methods, and an improvement in accuracy (92.10%) and sensitivity (94.59%) and a drastic reduction in false negatives are observed with the Cleveland dataset.