An efficient hybrid clustering algorithm for segmentation: Autocluster

A new automatic clustering algorithm has been proposed in this paper, which does not need clustering information, such as number of clusters and radius of density. Moreover, this algorithm generates robust results, and named Autocluster. Autocluster is a suitable algorithm for customer segmentation, and as it is known, clustering information is not available properly in customer segmentation. Autocluster applies concepts of partitioning clustering algorithms, hierarchical clustering algorithms and density-based clustering algorithm. Consequently, a new, automatic and high-precision algorithm has been proposed. Autocluster consists of four steps: developing 'distance matrix', identifying 'best point (data record)', developing 'point matrix' and 'clustering'. These steps have been explained comprehensively in this paper. Furthermore, iris database and a synthetic dataset has been analysed by Autocluster to verify its capabilities vs. K-means algorithm. Moreover, an Iranian insurance dataset has been clustered by Autocluster, which has shown satisfying results, compared to the results from K-means.