Model Selection Strategies for Determining the Optimal Number of Overlapping Clusters in Additive Overlapping Partitional Clustering

Abstract In various scientific fields, researchers make use of partitioning methods (e.g., K -means) to disclose the structural mechanisms underlying object by variable data. some instances, however, a grouping objects into clusters that are allowed overlap (i.e., assigning multiple clusters) might lead better representation clustering structure. To obtain an overlapping from data, Mirkin’s ADditive PROfile CLUStering (ADPROCLUS) model may be used. A major challenge when performing ADPROCLUS is determine optimal number which pertains selection problem. Up now, this problem has not been systematically investigated and almost no guidelines can found in literature regarding appropriate strategies for ADPROCLUS. Therefore, paper, several existing -means (a.o., CHull, Caliński-Harabasz, Krzanowski-Lai, Average Silhouette Width Dunn Index information-theoretic measures like AIC BIC) two cross-validation based tailored towards context compared each other extensive simulation study. The results demonstrate CHull outperforms all especially negative log-likelihood, associated with minimal stochastic extension ADPROCLUS, used as (mis)fit measure. analysis post hoc AIC-based strategy revealed performance obtained different—more appropriate—definition complexity