Whale Optimization Algorithm for Data Clustering

Abstract

Issues relating to clustering today are computational techniques, optimization, and performance of clustering algorithms. In this research, a metaheuristic grouping method of Whale Optimization Algorithm with modification to weight changes based on the food hunting method and behavior of humpback whales was proposed. A total of ten datasets obtained from the learning repository of UCI machine were used to evaluate the performance of the proposed algorithm compared to other clustering algorithms. Whale Optimization Algorithm can provide optimal solutions and more stable clustering results because there is no dependence on initial cluster center initialization. Moreover, clustering using Whale Optimization Algorithm produces clusters that are better than clusters using Ant Colony Optimization, Isodata, and Forgy. This can be seen from silhouette coefficients, time examination, variance examination and sum of square error of a number of test datasets obtained. Overall, results show that clusters of Whale Optimization Algorithm and the other three methods have nearly identical variance, and each cluster produces high intra-class similarity and low inter-class similarity.