Clustering with Mixed-type Attributes

• Activity Number: 520
• Type: Contributed
• Date/Time: Thursday, August 11, 2005 : 10:30 AM to 12:20 PM
• Sponsor: General Methodology
• Author(s): Jong-Min Kim*+ and Seoung-San Chae and William D. Warde
• Companies: University of Minnesota and Daejeon University and Oklahoma State University
• Keywords: Agglomerative Clustering Algorithm ; Rand's C Statistic ; Mixed-type Objects ; Association Coefficient

Abstract

Agglomerative clustering algorithms are well known in clustering sets of data. However, working on numeric values prohibits it from being used to cluster real-world data containing categorical values. In this research, we present sets of agglomerative clustering algorithms that extend to categorical domains and domains with mixed numeric and categorical values. Simple, matching Jaccard and Yule coefficients are used to measure similarity between objects. Similarities are converted to dissimilarities by using the formula dij = (1 - Sij )^(1/2), where Sij is the differently defined similarity coefficient between i-th and j-th objects. We use a dataset to demonstrate the clustering performance of clustering algorithms with different similarity measures. Our experiments on real-world datasets show agglomerative clustering algorithms are efficient when sets of data with mixed numeric and categorical attributes are clustered. Rand's (1971) C statistic is used to measure the retrieval ability of the agglomerative clustering algorithm with different types of calculating dissimilarity between objects.