Binary Whale Optimization Algorithm with Logarithmic Decreasing Time-Varying Modified Sigmoid Transfer Function for Descriptor Selection Problem

Abstract

In cheminformatics, choosing the right descriptors is a crucial step in improving predictive models, particularly those that use machine learning algorithms. Recently, researchers in cheminformatics have been lured to swarm intelligence to optimize the process of discovering relevant descriptors in the wrapper feature selection. This work introduced a new Binary Whale Optimization Algorithm, which utilized a novel time-varying modified Sigmoid transfer function with a modified logarithmic decreasing time-varying update strategy to improve the balancing of exploration and exploitation in WOA. The new Binary Whale Optimization Algorithm is integrated with wrapper feature selection and validated on descriptor selection problem to improve Amphetamine-type stimulants drug classification result. The suggested approach is compared to well-known swarm intelligence algorithms, and the results demonstrate its superiority.