Function approximations by coupling neural networks and genetic programming trees with oblique decision trees

This paper is concerning the development of the hybrid system of neural net works and genetic programming (GP) trees for problem domains where a comple te input space can be decomposed into several different subregions, and the se are well represented in the form of oblique decision tree. The overall a rchitecture of this system, called federated agents, consists of a facilita tor, local agents, and boundary agents. Neural networks are used as local a gents, each of which is expert at different subregions. GP trees serve as b oundary agents. A boundary agent refers to the one that specializes at only the borders of subregions where discontinuities or a few different pattern s may coexist. The facilitator is responsible for choosing the local agent that is suitable for given input data using the information obtained from o blique decision tree. However, there is a large possibility of selecting th e invalid local agent as result of the incorrect prediction of decision tre e, provided that input data is close enough to the boundaries. Such a situa tion can lead the federated agents to produce a higher prediction error tha n that of a single neural network trained over the whole input space. To de al with this, the approach taken in this paper is that the facilitator sele cts the boundary agent instead of the local agent when input data is closel y located at certain border of subregions. In this way, even if decision tr ee yields an incorrect prediction, the performance of the system is less af fected by it. The validity of our approach is examined by applying federate d agents to the approximation of the function with discontinuities and the configuration of the midship section of bulk cargo ships. (C) 1999 Elsevier Science Ltd. All rights reserved.