A novel IMW-mrmr feature selection approach for early detection of hypertension with HRV analysis

In the absence of control measures, hypertension poses a major threat to the world’s population’s health and can quickly develop into vascular disorders that are fatal. An estimated 1 in 13 persons worldwide suffer from a heart or circulatory ailment. The World Health Organisation (WHO) estimates that heart and circulatory disorders account for one in three fatalities worldwide. In the present world, early detection of high-risk patients is crucial due to the prevalence of vascular disorders and associated lethality. Heart rate variability (HRV) analysis is a predictive tool that can be used to identify the characteristics of patients who are at risk for unexpected cardiac deaths because it is a reliable predictor of these happenings. The challenge here is to detect the minute difference in HRV of high risk and low risk patients. The current approaches to this task rely solely on either Electrocardiogram (ECG) image data or HRV data and do not find effective ways to produce greater efficiency. With this challenge as a motivation, we propose an approach based on analysis of HRV features to assort the hypertensive and non-hypertensive patients. The proposed HRV analysis method is 2 step process—(1) The ECG signals of hypertensive patients are converted to RR intervals and (2) The HRV features are extracted from these RR. The features extracted are shortlisted by proposed Improved Mutable Weighted Maximum Relevance & Minimum Redundancy (IMW-mrmr) feature selection algorithm for an efficient classifier modelling. A comparative performance evaluation of three models—Neural Networks, Stacking ensemble, Bagging built using IMW-mrmr with other feature section algorithms shows a superior result by proposed algorithm with performance metrics F1 score (93.6), accuracy (94.6%), precision (94.2), recall (92.95) and AUC-ROC score(87.34).