Understanding and Enhancement of Internal Clustering Validation Measures

Abstract

Clustering validation has long been recognized as one of the vital issues essential to the success of clustering applications. In general, clustering validation can be categorized into two classes, external clustering validation and internal clustering validation. In this paper, we focus on internal clustering validation and present a study of 11 widely used internal clustering validation measures for crisp clustering. The results of this study indicate that these existing measures have certain limitations in different application scenarios. As an alternative choice, we propose a new internal clustering validation measure, named clustering validation index based on nearest neighbors (CVNN), which is based on the notion of nearest neighbors. This measure can dynamically select multiple objects as representatives for different clusters in different situations. Experimental results show that CVNN outperforms the existing measures on both synthetic data and real-world data in different application scenarios.

Architecture of the Mouse Brain Synaptome

“…This gave rise to a comprehensive pool of clustering that contained 2400 different clustering results on the same dataset. Each clustering result was then quality evaluated based on the 11 clustering validity indices (Liu et al., 2013) in machine learning. These indices were finally used as weights to combine 2400 clustering results: results with higher validity indexes were given more priority in the final clustering result, and the cluster number was selected from results of the highest validity indices.…”

Rethinking the Identification of Urban Centers from the Perspective of Function Distribution: A Framework Based on Point-of-Interest Data

“…After performing the DBSCAN algorithm, the effectiveness of UC identification was evaluated [49][50][51]. Consistent with the study conducted by Liu, Li, Xiong, Gao, Wu, and Wu [51], CVNN was employed to compare the different versions of UC identification with diverse k-values. Typically, the best version of UC identification has the lowest CVNN value.…”

Cap Rate as the Interpretative Variable of the Urban Real Estate Capital Asset: A Comparison of Different Sub-Market Definitions in Palermo, Italy

“…Since the number of possible partitions is high-e.g., for n = 20, p is 524.287-it is recommended to reduce the number by selecting a number of groups as appropriate. The Calinski-Harabasz index, CH [48,49], identifies the best subdivision, i.e., the one that maximises the external heterogeneity between the data groups and minimises the internal one.…”