Evolutionary multi-objective clustering with adaptive local search

Abstract

In many real-world applications, the accurate number of clusters in the data set may be unknown in advance. In addition, clustering criteria are usually high dimensional, nonlinear and multi-model functions and most existing clustering algorithms are only able to achieve a clustering solution that locally optimizes them. Therefore, a single clustering criterion sometimes fails to identify all clusters in a data set successfully. This paper presents a novel multi-objective evolutionary clustering algorithm based on adaptive local search that mitigates the above disadvantages of existing clustering algorithms. Unlike the conventional local search, the proposed adaptive local search scheme automatically determines whether local search is used in an evolutionary cycle or not. Experimental results on several artificial and real data sets demonstrate that the proposed algorithm can identify the accurate number of clusters in the data sets automatically and simultaneously achieves a high quality clustering solution. The superiority of the proposed algorithm over some single-objective clustering algorithms and existing multi-objective evolutionary clustering algorithms is also confirmed by the experimental results.