A Binary Walrus Optimizer for Feature Selection Problem

Abstract

Feature selection is a very crucial step in machine learning. It plays an important role for enhancing model performance, interpretability and efficiency. The main goal of feature selection is to select the optimal subset of features which is considered as the relevant features. In this paper, we introduce a novel feature selection approach based on a novel optimization approach called Walrus Optimization Algorithm (WOA) proposed by Muxuan Han et al. in 2023 and never tested in the context of feature selection. A new binary version of Walrus Optimization is proposed and adapted to the problem of feature selection. The fitness function is composed of three important terms: classification accuracy rate, correlation and class separability measure based on Jaccard Index. To evaluate the performance of the proposed approach, five synthetic datasets was used: CorrAl, m-of-n-3-7-10, Monk1, Monk2 and Monk3. In addition, the approach was tested on real-world DNA microarray datasets: colon cancer, leukemia, breast cancer ovarian cancer, lung cancer and DLBC cancer (diffuse large B-cell lymphoma). The results demonstrate that the proposed can produce a high classification accuracy rate and a good diagnostic of cancer.