Classification Prediction of Familial Hypercholesterolemia using Ensemble-based Classifier with Feature Selection and Rebalancing Technique

Abstract

Familial hypercholesterolemia (FH) is the most prevalent hereditary hyperlipidemia. Although FH is a significant risk factor of premature coronary heart disease (CHD), it is treatable if detected early and prompt intervention is given. Nevertheless, most people with FH receive inadequate diagnosis and treatment, which results in missed opportunities for premature CHD prevention. Therefore, an efficient and faster method of diagnosing FH is crucial for early identification among Malaysians, especially in this age of technology. This study aims to evaluate the performance of ensemble-based classifier and rebalancing strategy with Synthetic Minority Oversampling Technique (SMOTE) towards FH diagnosis in the Malaysian population. Our proposed ensemble-based classifier consists of a combination decision tree, random forest, extreme gradient boosting, ensemble-based classifier using majority voting technique. We also applied Recursive Feature Elimination (RFE) to identify significant features across three well-known diagnostic tools. Experimental findings demonstrate that our proposed ensembled-based classifier with RFE and SMOTE, considerably outperforms the baseline by 99.32% in terms of accuracy, precision, recall, micro-average, macro-average, and G-mean. The proposed ensemble-based classifier with RFE approach selected the same significant features of FH for each of the three diagnostic criteria. We hope that the ensemble-based classifier will aid early detection of FH among Malaysian population and can be used as predictive tool for future studies.