Clustering mixed type attributes in large dataset

Abstract

Clustering is a widely used technique in data mining, now there exists many clustering algorithms, but most existing clustering algorithms either are limited to handle the single attribute or can handle both data types but are not efficient when clustering large data sets. Few algorithms can do both well. In this paper, we propose a clustering algorithm CFIKP that can handle large datasets with mixed type of attributes. We first use CF*-tree to pre-cluster datasets. After the dense regions are stored in leaf nodes, then we look every dense region as a single point and use an improved k-prototype to cluster such dense regions. Experiments show that the CFIKP algorithm is very efficient in clustering large datasets with mixed type of attributes. © Springer-Verlag Berlin Heidelberg 2005.