Linear and non-linear HRV features for the prediction of heart disease among smokers: a predictive evaluation of machine learning model

Abstract

In smokers it is found that they have increased sympathetic and reduced parasympathetic activity in the heart rate variability (HRV) analysis by researchers. Smoking lowers HRV, it harms heart health function as it influences the Autonomous Nervous System (ANS). Therefore, HRV features are useful for the prediction of heart disease. Due to poor health awareness and inadequate lifestyle, heart patients are proliferating. Hence, having a model that can easily recognize the predominance of heart disease potentially is essential. In this study, an HRV parameter is used to assess the potential of four machine learning techniques to predict heart disease among smokers. These techniques assessed on classification indices of accuracy, precision, sensitivity, specificity, misclassification rate, F1 score, area under the curve (AUC), kappa value, and mean square error (MSE). These techniques also evaluated on the receiver operating characteristic curve (ROC). The final model has the highest classification accuracy of 0.94 which was reported using a k-nearest neighbors (k-NN) method with 0.95, 0.92, 0.96, 0.94, 0.061, 0.878, 0.95 and 0.246 precision, sensitivity, specificity, F1 score, kappa value, AUC, and MSE respectively.