Clustering Based on Cuckoo Optimization Algorithm

This paper presents four novel clustering methods based on a recent powerful evolutionary algorithm called Cuckoo Optimization Algorithm (COA) inspired by nesting behavior and immigration of cuckoo birds. To take advantage of COA in clustering, here, an individual cuckoo represents a candidate solution consisting of clusters’ centroids. Fitness function calculates sum of intra cluster distances. Three proposed approaches named Random COA Clustering, Chaotic COA Clustering and K-means COA Clustering differ in initial step of original COA algorithm. In COA Clustering, initial population is produced randomly. In Chaotic COA Clustering, to cover whole search space and enrich algorithm, chaotic Arnold’s Cat map is used to produce initial population instead of randomness. In Kmeans COA Clustering, to start from closer to global optimum, well-known K-means algorithm is conducted to produce initial cuckoos. In order to local search in COA, each cuckoo lays its own eggs within a specific radius. The aim of producing better neighbors and escape local optimum in proposed Enhanced COA Clustering (ECOAC), this boundary doesn’t exist and each cuckoo puts its eggs via Lévy flight. The results of conducting these novel methods on four UCI datasets illustrate their comparable stability and power of them.