An Interactive Approach to Building Classiffication Models by Clustering and Cluster Validation

Abstract

This paper presents the decision clusters classifier (DCC) for database mining. A DCC model consists of a small set of decision clusters extracted from a tree of clusters generated by a clustering algorithm from the training data set. A decision cluster is associated to one of the classes in the data set and used to determine the class of new objects. A DCC model classifies new objects by deciding which decision clusters these objects belong to. In making classification decisions, DCC is similar to the k-nearest neighbor classification scheme but its model building process is different. In this paper, we describe an interactive approach to building DCC models by stepwise clustering the training data set and validating the clusters using data visualization techniques. Our initial results on some public benchmarking data sets have shown that DCC models outperform the some existing popular classification methods.