Hybrid Feature Subset Selection Approach for Fuzzy-Extreme Learning Machine

Abstract

Extreme learning machine (ELM) is a fast learning algorithm which has various advancements in terms of speed, reliability and generalized performance. Perhaps, it lacks the ability to solve the weighted classification problem. One of the alternative solutions to this problem is fuzzy-ELM (F-ELM) which combines the advantages of a fuzzy system and ELM. The major aim of this research is to handle the feature subset selection (FSS) and weighted classification problem. In this paper, a hybrid FSS approach for F-ELM classifier (H-FELM) is designed for binary and multiclass classification on the clinical dataset. Experimental results illustrate that H-FELM has the capability to handle these problems with improved classification accuracy by selecting prominent features which are relevant and non-redundant. The selection of such features can reduce the computational overhead and hence minimize the learning time. In order to validate the efficiency and effectiveness of the proposed algorithm, the comparative performance is evaluated by using four different strategies—1. existing F-ELM, 2. existing F-ELM with FSS, 3. ELM and F-ELM and 4. ELM with FSS and F-ELM with FSS. It is observed that the designed algorithm achieves an improvement of 9–10% in accuracy over existing results with almost reduction of 50% features for the clinical dataset.