Evolving neural networks in classification.

摘要

This research is to develop a biologically inspired hybrid intelligent system—evolving neural networks—that can be used in data mining, especially in classification problems. This hybrid system employs computational intelligence methodologies, such as neural networks and genetic algorithms. Genetic algorithms have been applied to the automatic generation of neural networks in which the network is evaluated on data collected from the environment. The proposed system is able to select inputs from the environment and control its topology. This ability enables the system to improve generalization capability and adapt successfully to a given problem. Also, the user-defined objective function offers customized classification. Recently, the problems with a single network have motivated the biologically inspired neural models, such as combining multiple networks. In this sense, an ensemble of evolving neural networks is also proposed. The proposed system is tested by various data sets and produces better performance than both classical neural networks and simple ensemble methods. The achievement of this research is the development of the smart architecture for engineering systems that can adapt to the environment by its own ability. This architecture can be applied to business intelligence for identifying characteristics of data, manufacturing for pattern recognition, medical areas for analyzing diseases, and others.