GENEmops: Supervised feature selection from high dimensional biomedical dataset

Identification of differentially expressed genes, lying beneath the carcinogenic expression, is still very crucial for accurate detection and treatment of the disease. The challenge of a large number of attributes compared to a small number of instances and the prediction of highly discriminative genes requires an effective method. It can be regarded as a multi-objective problem that involves minimization of the number of selected genes and maximization of the classification performance. It is expected to find the optimal count of the most significant genes which are strongly associated with the classification of cancer. In this paper, we have proposed a framework entitled as GENEmops for gene selection and subsequent classification of the disease. The core of the GENEmops is inspired by multi-objective player selection strategy based hybrid population search (MOPS-HPS). The proposed system uses a multi-filtering and adaptive parameter tuning approach for gene selection. A new graded rotational blending operator is introduced to enhance the exploitation capability of the hybrid wrapper based scheme. Unlike the most current existing methods which employ some strategy to transmute the continuous search space to binary search space, it uses an adaptive way for binary conversion which is stochastically updated during the search phase. GENEmops also improves the performance of the classifier by tuning its parameters. The efficiency of the proposed GENEmops is tested on sixteen biological datasets (eight binary class and eight multi-class) and compared with state-of-the-art computationally intelligent multi-objective approaches. Experimental results demonstrate the efficiency of the proposed work. (C) 2022 Elsevier B.V. All rights reserved.