Novel Auxiliary Techniques in Clustering

Clustering is grouping of patterns according to similarity or distance in different perspectives. Various data representations, similarity measurements and organization manners are led to several classes of clustering methods. In this paper a new combinatorial method is proposed that iteratively uses another clustering method such as Rival Penalized Competitive Learning (RPCL) or K-Means as the core of clustering system. Moreover, some novel auxiliary techniques are suggested to increase the clustering performance. The proposed method has been compared with well known clustering methods such as K-Means, its improvement, ISODATA and DSRPCL2. The new combinatorial technique can detect the drawbacks of core clustering method and improve its efficiency. Our method is applied on some standard multi-class datasets. After clustering, labels of grouped samples in each cluster are compared with their real class labels to show the accuracy of clustering.