A Novel Hybridization of Self-adaptive Multi-verse Optimizer over K-Means for Data Clustering

Abstract

Although clustering algorithms are a popular way to find the relationship among a collection of data, these algorithms have to deal with various types of challenges such as slow convergence rate, converging to local optima, and requiring the number of clusters in advance. In order to solve these drawbacks in one of the most famous clustering algorithms called K-Means, this paper has been presented a novel method based on nature-inspired algorithms in combination with clustering technique to construct a hybrid method for solving the clustering as an optimization problem in a reasonable time. Many nature-inspired algorithms have been successfully used to solve non-linear optimization problems. This paper proposed an algorithm which uses a recently introduced nature-inspired algorithm called Multi- Verse Optimizer (MVO) over K-Means to minimize the cluster integrity and maximize the distance between clusters by finding the optimal number of clusters (K) as well as initial centroid for clusters. The proposed method has been tested using ten datasets and compared with Particle Swarm Optimization (PSO), Genetic Algorithm (GA), K-Means with random initial centroids, and K-Means++ to show the considerable improvement of clustering by using the proposed method. The results have shown that our new self-adaptive method outperforms other comparing nature-inspired algorithms both in cluster integrity and convergence rate.