An Intelligent System Based on Heart Rate Variability Measures and Machine Learning Techniques for Classification of Normal and Growth Restricted Children

Abstract

Growth restricted children have higher predisposition of developing metabolic syndrome, type-2 diabetes, hypertension and cardiovascular problems in later life. Numerous intelligence systems that have proved their effectiveness for detection of cardiac abnormalities to support medical diagnosis. Previous studies used heart rate variability (HRV) analysis techniques for distinguishing normal and growth restricted children, however those studies did not use intelligent systems for this purpose. The aim of present study is to develop an intelligent system using HRV analysis measures and machine learning (ML) techniques for early detection of cardiac abnormalities in growth restricted children. We performed two sets of experiments using interbeat interval time series data of the normal and growth restricted children and different combinations of individual characteristics of the subjects. Several ML algorithms such as linear discriminant analysis (LDA), support vector machine with linear and sigmoid kernels (SVML and SVMS), random forest (RF), and RPart are used for developing intelligent system to classify normal and growth restricted children. We evaluated the performance of the classifiers using sensitivity, specificity, area under receiver operator characteristic curve and total accuracy. The results reveal that the LDA is robust for classifying normal and LBW-IUGR children with 100% accuracy at all cross validation formulations. The SVMS and LDA revealed highest accuracy, whereas, RF and Rpart were robust for classifying LBW-IUGR and ABW_IUGR. Our findings show that the intelligent system developed using HRV analysis markers and ML techniques could be a reliable tool for identifying future risk of cardiac abnormalities in IUGR children.