Evolutionary and Swarm Intelligence Methods for Partitional Hard Clustering

Abstract

Clustering is an unsupervised classification method where objects in the unlabeled data set are classified on the basis of some similarity measure. The conventional partitional clustering algorithms, e.g., K-Means, K-Medoids have several disadvantages such as the final solution is dependent on initial solution, they easily stuck into local optima. The nature inspired population based global search optimization methods offer to be more effective to overcome the deficiencies of the conventional partitional clustering methods as they possess several desired key features like up gradation of the candidate solutions iteratively, decentralization, parallel nature, and self organizing behavior. In this work, we compare the performance of widely applied evolutionary algorithms namely Genetic Algorithm (GA) and Differential Evolution (DE), and swarm intelligence methods namely Particle Swarm Optimization (PSO) and Artificial Bee Colony (ABC) to find the clustering solutions by evaluating the quality of cluster with internal validity criteria, Sum of Square Error (SSE), which is based on compactness of cluster. Extensive results are compared based on three real and one synthetic data sets.