Hybrid Binary Grey Wolf with Jaya Optimizer for Biomarker Selection from Cancer Datasets

Abstract

GWO (Grey Wolf optimizer) is the leading metaheuristic algorithm preferred by most researchers to solve different feature selection and optimization problems. Stagnation in local optima is a major concern that exists and it affects the performance of the machine-learning model. Exploration (search field) and exploitation (identifying optimal solutions) are the two most vital concepts in each metaheuristic algorithm. Properly balancing both to achieve a better result is a critical task as GWO is stuck in local optima. This paper presents a hybrid GWO with the Jaya algorithm (JA) as a local search (exploitation) to solve stagnation issues. Most of these methods are designed for solving complex continuous problems. A sigmoid transfer function is used to convert continuous search space to binary, which creates an environment for feature selection. The high-dimensional datasets are used to evaluate the most prominent features. To evaluate the performance of the proposed model, three different cancer datasets were used. The performance of the proposed model is compared with different state-of-the-art machine learning models. The analysis of the results shows that GWO optimized with local search (Jaya) performs better in terms of accuracy, feature size (selected), and computation time.