A neural network classification tree
Yi‐Shiou Chen, Tah‐Hsiung Chu · 2002
Neural network and classification tree are two basic approaches for pattern classification. However, neural network requires massive parallel processing, and classification tree suffers overlapped nodes and error accumulation due to its crisp partition. In this paper, we integrate neural network, classification tree and intelligent search strategy to develop a novel neural network model, called neural network classification tree (NNCT), to reduce the computational complexity without sacrificing classification accuracy. NNCT has a tree structure, and each node contains a special neural network, called hint perceptron, to provide information for efficient search algorithm. In NNCT, we propose a definition of admissibility and a rule of maximizing dissimilarity to train the hint perceptrons with training samples. Simulation results show that NNCT, without degrading classification accuracy, has much less computational complexity than the regular neural networks.