Improving Heart Disease Prediction of Classifiers with Data Transformation using PCA and Relief Feature Selection

Abstract

Cardiovascular disorders (CVD) are the key cause of mortality worldwide. One in three male premature deaths and one in five female premature deaths are thought to be attributable to Cardiovascular disorders. Early prediction of CVDs may help to attenuate the disease, potentially lowering death rates. The existence of cardiac disease can be predicted using machine learning approaches; however, the effectiveness of the classifiers may be enhanced by applying PCA, relief feature selection, and data transformation techniques. The objective of employing data transformation, PCA, and relief feature selection approaches is to enhance classifier performance and increase the interpretability and ability of classifiers to predict heart disease. Heart disease anticipating is a challenging problem in the field of healthcare. This uses popular supervised machine learning (ML) algorithms including k-NN, LR, DT, RF, SVM, and ANN to help healthcare practitioners and specialists easily identify the prevalence of heart-related illnesses in patients. In these trials, data transformation is achieved using PCA, normalized features, and relief techniques, and RF surpasses all other classifiers with a prediction accuracy of 90%, followed by ANN and DT with AUCs of 87% and 86%, respectively. SVM and Naive Bayes classifiers were shown to be lesser effective at predicting heart disease.