A new internal index based on density core for clustering validation

Abstract Clustering validation which is applied to the evaluation of clustering results has been recognized as one of the vital issues for clustering application. Density core, a set of connected maximum local density peaks, can approximately represent the structure of a cluster without being affected by noises. Therefore, a density-core-based clustering validation index (DCVI) with minimum spanning tree (MST) is proposed to solve the problem that noises and arbitrary shapes have great influence on the performance of widely used internal validation measures. As for data sets containing arbitrarily shaped clusters with noises, DCVI can effectively obtain the optimal number of clusters. Moreover, experimental results of both synthetic and real data sets indicate that DCVI outperforms most existing measures.