Decision Tree classifier using theme based partitioning

Abstract

Decision Tree (DT) is one of the widely adopted non-metric classification techniques in pattern recognition, data mining and related areas. With the increase in dimensionality of the data, the classical decision tree techniques may not exhibit higher classification rate due to curse of dimensionality phenomenon. In this paper, we propose a partitioning based Decision Tree method which creates sub-objects for each data object based on themes, constructs multiple local decision trees using the sub-objects, and combines the decisions based on nearest neighbour rule. Our empirical results on Teacher data sets confirm the improved classification rate of the proposed method over other decision tree classifiers (CART, C4.5, C5.0).