Extracting Classification Rules with Support Rough Neural Networks

Abstract

Classification is an important theme in data mining. Rough sets and neural networks are two technologies frequently applied to data mining tasks. Integrating the advantages of two approaches, this paper presents a hybrid system to extract efficiently classification rules from a decision table. The neural network system and rough set theory are completely integrated to into a hybrid system and use cooperatively for classification support. Through rough set approach a decision table is first reduced by removing redundant attributes without any classification information loss. Then a rough neural network is trained to extract the rules set form the reduced decision table. Finally, classification rules are generated from the reduced decision table by rough neural network. In addition, a new algorithm of finding a reduct and a new algorithm of rule generation from a decision table are also proposed. The effectiveness of our approach is verified by the experiments comparing with traditional rough set approach.