A Point Symmetry-Based Automatic Clustering Approach Using Differential Evolution

Clustering is a core problem in data mining and machine learning though it is widely applied in many fields. Recently, it is very popular to use the evolutionary algorithm to solve the problem. This paper proposes an automatic clustering differential evolution (DE) technique for the problem. This approach can be characterized by (i) proposing a modified point symmetry-based cluster validity index (CVI) as a measure of the validity of the corresponding partitioning, (ii) using the Kd-tree based nearest neighbor search to reduce the complexity of finding the closest symmetric point, and (iii) employing a new representation to represent an individual. Experiments conducted on 6 artificial data sets of diverse complexities indicate that this approach is suitable for both the symmetrical intra-clusters and the symmetrical inter-clusters. In addition, it is able to find the optimal number of clusters of the data. Furthermore, based on the comparison with the original point symmetry-based CVI, this proposed point symmetry-based CVI shows better performance in terms of the F-measure and the number of clusters found.