Hybrid decision tree based on inferred attribute

Abstract

This paper proposes a novel hybrid approach based on Regression, Rough sets and a Classifier aiming at effective dimensionality reduction henceforth a classifier with less time complexity. It focuses on dimension enhancement by embedding Inferred Attribute into the decision table, named as Augmented Decision Table (ADT). An effective minimal set of attributes (reduct) are derived by employing Rough set theory on ADT. The projected ADT based on the reduct is used for building a classifier. The philosophy is demonstrated by considering popular functional forms for Inferred Attribute like Linear Regression and Quadratic Regression and by adopting CART as the classifier. It is observed that inclusion of Inferred Attribute reduces the size of the reduct significantly without compromising the accuracy. However improvement in classification time depends on the choice of the Inferred Attribute model form. © 2010 ACM.