A Robust Intuitionistic Fuzzy Constraint Score based Potential Feature Subset Selection for Chronic Diseases Detection

Abstract

This work proposes a novel feature selection algorithm for high-dimensional features in real-time datasets for prediction or classification. Conventional methods assume dataset values as crisp formats, but in real datasets, instances are represented in linguistic formats, requiring the use of uncertainty theories. The Intuitionistic Fuzzy Similarity based constraint score is proposed, where each feature is denoted as an independent variable and the class variable as a dependent variable. The features are represented in triplet form, with grade of belongingness, non-belongingness, and hesitancy index to maximize relevancy and reduce redundancy. Pairwise similarity matching is computed using Intuitionistic fuzzy similarity distance measure for supervised learning and intuitionistic fuzzy K-NN for semi-supervised learning. Potential feature subsets are selected and validated using deep learning algorithms. The results show that the proposed Intuitionistic fuzzy Constraint score feature selection algorithm produces optimal results compared to other state-of-the-art methods in chronic disease prediction.