Kernel-based decision cluster classifier

The decision cluster classifier (DCC) is represented as a tree-like classifier. A cluster tree is built with a clustering method. A new object is classified according to the distance between the object and the cluster-nodes of the cluster tree. In this paper, we propose a Kernel distance to replace the Euclidean distance in DCC classifier and define a Kernel-based decision classifier (KDCC). We formulate the classification model of DCC as space partition by Voronoi diagram in which the set of cells is the set of decision clusters of DCC. To classify a new object is to find the cell the object falls through a distance measure. We observe that the sizes and densities of decision cluster cells are unbalanced, which results in high misclassification rate with Euclidean distance. The misclassification rate can be reduced by Kernel distance. Experiment results have shown that KDCC outperformed DCC or some other classifiers on complex data sets. ICIC International © 2010 ISSN 1881-803X.