Soft Set-based Parameter Reduction Algorithm Through a Discernibility Matrix and the Hybrid Approach for the Risk-Factor Prediction of Cardiovascular Diseases by Various Machine Learning Techniques

Abstract

Parameter reduction without performance degradation is a promising task in decision-making problems. For instance, a great challenge exists in constructing cost functions in gaming theory. Nevertheless, soft set theory handles all its drawbacks conveniently through a new tool for the choice function mathematically. In this paper, we propose an algorithm (SSPRDM) for parameter reduction of soft sets through discernibility matrices, and it is implemented in detecting the risk factor of heart disease problems by using six types of machine learning techniques. The parameters are extracted from the heart disease patient data by the SSPRDM algorithm, and then six machine learning techniques (LDA, KNN, SVM, CART, NB, RF) are performed in the prediction of risk factors for heart disease. The experimental results showed that the present hybrid approach provides an accuracy of 88.46% in the Random Forest technique, whereas the same Random Forest classifier provides an accuracy of 69.23% in the prediction of risk factors of cardiovascular disease (CVD) diagnosis in the earlier work which is a drastic improvement. Moreover, out of 18 parameter reductions, the core component is identified as Total Cholesterol, which is to be considered in all types of CVD diagnosis, whereas Sugar-Fasting (C), Total-Cholesterol (G), and HDL-Cholesterol (I) are the core components identified in three parameter reductions ABCEGHI, ACFGIJ, and BCFGIJK.