Improved Water Cycle Algorithm and K-Means Based Method for Data Clustering

Abstract

K-means is a classical clustering method, but it is easy to fall into local optimums because of poor centers. Inspired by the good global search performance of Inter-Peer Communication Mechanism Based Water Cycle Algorithm (IPCWCA), three hybrid methods based on IPCWCA and K-means are presented in this paper, which are used to address the shortcoming of K-means and explore better clustering approaches. The hybrid methods consist of two modules successively: IPCWCA module and K-means module, which means that K-means module will inherit the best individual from IPCWCA module to start its clustering process. Compared with original K-means and WCA + K-means methods on eight datasets (including two customer segmentation datasets) based on SSE, accuracy and Friedman test, proposed methods show greater potential to solve clustering problems both in simple and customer segmentation datasets.