A Multicriteria Decision Making Approach for Estimating the Number of Clusters in a Data Set

• Published in: PloS one Vol. 7; no. 7; p. e41713
• Main Authors: Peng, Yi, Zhang, Yong, Kou, Gang, Shi, Yong
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 27.07.2012; Public Library of Science (PLoS)
• Subjects: Algorithms; Analysis; Artificial Intelligence; Breast cancer; Cluster Analysis; Clustering; Clusters; Computer Science; Data mining; Datasets; Decision Making; Engineering; Estimation; Forensic sciences; Information technology; Knowledge discovery; Learning algorithms; Linear programming; Localization; Machine learning; Mathematics; Methods; Multiple criterion; Social and Behavioral Sciences; Studies; Validity; China
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0041713

Determining the number of clusters in a data set is an essential yet difficult step in cluster analysis. Since this task involves more than one criterion, it can be modeled as a multiple criteria decision making (MCDM) problem. This paper proposes a multiple criteria decision making (MCDM)-based approach to estimate the number of clusters for a given data set. In this approach, MCDM methods consider different numbers of clusters as alternatives and the outputs of any clustering algorithm on validity measures as criteria. The proposed method is examined by an experimental study using three MCDM methods, the well-known clustering algorithm--k-means, ten relative measures, and fifteen public-domain UCI machine learning data sets. The results show that MCDM methods work fairly well in estimating the number of clusters in the data and outperform the ten relative measures considered in the study.