Hybrid knowledge acquisition by integrating decision trees and neural networks

Decision tree induction is one of the most effective techniques for acquiring classification knowledge. However, appropriate pre- and post-processors have to be prepared to achieve continuous input/output mapping, because the decision trees basically deal with symbolic knowledge. On the other hand, an artificial neural network is suitable for such a purpose, however, its initial structure is difficult to constitute. The authors´ research goal is to develop a sophisticated knowledge acquisition system integrating decision tree induction for identifying the fundamental structure of the knowledge and neural network generation for realizing an adaptive processor based on the knowledge structure obtained as a decision tree. This paper reports an experimental approach to this goal by constructing a neural network based on the result of decision tree induction from symbolic examples, and analyzing the network to elicit hidden knowledge in numerical examples