We present a novel fuzzy clustering technique called CRUDAW that allows a data miner to assign weights on the attributes of a data set based on their importance (to the data miner) for clustering. The technique uses a novel approach to select initial seeds deterministically (not randomly) using the density of the records of a data set. CRUDAW also selects the initial fuzzy membership degrees deterministically. Moreover, it uses a novel approach for measuring distance considering the user defined weights of the attributes. While measuring the distance between the values of a categorical attribute the technique takes the similarity of the values into consideration instead of considering the distance to be either 0 or 1. Complete algorithm for CRUDAW is presented in the paper. We experimentally compare our technique with a few existing techniques – namely SABC, GFCM, and KL-FCM-GM based on various evaluation criteria called Silhouette coefficient, F-measure, purity and entropy. We also use t-test, confidence interval test and time complexity in evaluating the performance of our technique. Four data sets available from UCI machine learning repository are used in the experiments. Our experimental results indicate that CRUDAW performs significantly better than the existing techniques in producing high quality clusters.
|Cite as: Rahman, M.A. and Islam, M.Z. (2012). CRUDAW: A Novel Fuzzy Technique for Clustering Records Following User Defined Attribute Weights. In Proc. Data Mining and Analytics 2012 (AusDM 2012) Sydney, Australia. CRPIT, 134. Zhao, Y., Li, J. , Kennedy, P.J. and Christen, P. Eds., ACS. 27 - 42 |
(local if available)