A Constructive Feature Grouping Approach for Analyzing the Feature Dominance to Predict Cardiovascular Disease

Abstract

Predicting the resistivity of an individual is essential for the optimal and prompt treatment against cardiovascular disease (CVD) in the earlier stage, which recommends the requirement for productive risk evaluation tools. The data-driven-based approach can predict every individual's risk by handling the crucial data patterns. To facilitate clinically applicable CVD prediction resolving the missing data patterns and interpretability issues using machine learning (ML) approaches. Here, a multi-tier model is proposed for mining missing data patterns. Initially, data fusion is adapted to describe the block-wise data patterns. It enables patient data (1) grouping-based feature learning and imputation of missing data and (2) prediction model considering the data availability. The feature selection process uses group characterization to uncover the risk factors. Then, the boosting model is generalized for identifying the patient's sub-group. The experimentation is done on an online available UCI ML dataset to demonstrate the significance of the model compared to various other approaches. The model attains 99% prediction accuracy, which is substantially higher than other approaches.