Research And Application On Artificial Neural Network Algorithms In Data Mining

Along with the application of information management system in different fields, the data are continuously stored in the database. Among them, people expect to find out potential knowledge that will help them make decisions. The emergence and development of data mining are just based on this expectation. As a midpoint of different subject studies, data mining involves a great deal of subjects such as database, statistic, machine learning, artificial intelligence, high performance computing, pattern recognition and data visualization etc. Among them, artificial neural network is widely used due to its ability of inherent non-linear processing, adaptive learning and high fault-tolerance.This paper will emphasize on the theory of back-propagation (BP) neural network and its application in the data mining on the basis of different neural network models. It is well known that there are three main inherent problems of being prone to getting stuck in a local minimum, the slow convergence speed and lack of stable generalization ability in BP network algorithm. So this paper will take more efforts on the improvement of algorithms and optimization of network structure onpoint of view of gradient descent, numerical optimization and combining other theory.According to the requirements which raise in the China Customs taxation evaluation, by the VC++.NET developing tool, this paper presents an improved learning algorithm which combines adding momentum term and self-adapting the learning rate. It is the first time that BP neural network model is introduced in the China Customs taxation. On the basis of this improved BP neural network, a new analyzing system of customs taxation is successfully set up. After testing of real tax data, it is demonstrated that this improved BP neural network algorithm has better performance on accuracy of evaluation than basic BP neural network. This improved model can be widely used in the near future.