Improving Amphetamine-Type Stimulants Drug Classification Using Binary Whale Optimization Algorithm as Relevant Descriptors Selection Technique

Swarm intelligence (SI) has become a popular choice to optimize the wrapper feature selection technique. It has attracted this research to employ a binary whale optimization algorithm (BWOA) to solve the molecular descriptors selection problem in ATS drugs classification. This effort is to enhance the learning and prediction ability of the classifier to generate good classification results. S-shaped transfer functions are adopted to generate BWOA, which are then consolidated in the wrapper feature selection with a k-Nearest Neighbor (k-NN) classifier. Our goal is to investigate the influence of different sigmoid transfer functions in BWOA on the selection of significant molecular descriptors and classification accuracy. Several metrics and Wilcoxon’s rank-sum test are utilized for performance evaluation. Experimental results reveal that the BWOA-S5 offers performance advantages with the lowest fitness value, fast convergence, high classification accuracy and, small feature subset. Furthermore, the generalization of the optimal molecular descriptor subset is ratified by six different classifiers.