A New Approach to Cohesion Measurement: Region-Based Clustering Validation

Abstract

Clustering assigns objects to clusters based on similarity, aiming to ensure that objects within the same cluster are similar and those in different clusters are dissimilar. Evaluating clustering quality is crucial and challenging. Thus, researchers have proposed clustering validation indices namely internal and external validation indices. Internal indices assess clustering quality using intrinsic information within a dataset. We focus on internal validation indices for their real-world applicability. In this paper, we have proposed a novel region-based internal validation (RCV) index. Our index incorporates the division of each cluster into three distinct regions which are the inner, middle, and outer regions. according to the clusters' center and their corresponding radius, we split each cluster into the aforementioned regions. The average distance is then computed for each region, and a penalty factor is applied to these average distances. By summing up the three penalized average distances, a Region Cluster Validation (RCV) score is obtained for each cluster. The RCV scores for all clusters are then summed together to yield an overall measure of cluster validity. A lower index value indicates better clustering quality. Experiment results on the synthetic and real-world datasets exhibit the usability and effectiveness RCV index.