Abstract: This study used an unsupervised approach based on an optimization algorithm to cluster individuals by demographic and personality-trait similarity, and the resulting clusters were then characterized in terms of predefined drug-use categories. The phase-angle-coded snake optimization algorithm (θ-SO) has been used to solve clustering problems, but it often suffers from unsatisfactory accuracy due to sensitivity to control parameters that constrain the exploration phase. To overcome this limitation, this study proposed a snake optimization algorithm for clustering (SOC) by presenting a new equation specifically designed to improve results by directing the search to efficient regions. The optimization performance of the proposed SOC algorithm is evaluated using 19 drug consumption datasets against the most competitive algorithms reported in recent studies include genetic algorithm, grey wolf optimizer for clustering, simulated annealing and θ-SO algorithm. The experiments were conducted using four metrics: minimum clustering distance (Intra-clustering distance), purity, rand and entropy. The results were compared with mentioned algorithms indicated that the proposed SOC algorithm outperformed them by 79% based on average minimum distance, 52% based on purity, 73% based on rand and 63% based on entropy. The output from this result can be used to improve the accuracy and stability of the clustering.

Keywords: Unsupervised learning, Drug consumption data, Behavioral data analysis, Drug diagnosis, Metaheuristic

Pages: 749-762