Abstract

Feature selection technique attempts to select and remove irrelevant features while ensuring that an informative subset of features remains in the dataset. The performance of a classifier often depends on the feature subset used for the robust classification task. In the medical and healthcare application domain, classification accuracy plays a vital role. The higher level of false negatives in medical diagnosis systems may raise the risk of patients not employing the necessary treatment they need. In this article, we have proposed an unsupervised feature selection method that underlines the concepts of rough set theory for the task of classification of high-dimensional datasets. Experiments are carried out on seven public domain healthcare and life science related datasets. The obtained experimental results justify the significance of the proposed method over five other state-of-the-art feature selection methods.