The Research Of The Data Mining Technique Based On Neural Network

Data mining is to locate covered , unknown knowledge and pattern potentially useful for strategic decision making from massive data . Those patterns contain the specific relationship among a set of objects in the data base, providing useful information and the basis for desion and forecasting .Due to the wide application of data mining, the nuclear algorithms used to solve different problems in different fields are diverse in nature. To deal with the complexity of data structure and the diversity of information encountered in education, we propose to solve those problems with NN algorithms. NN algorithms is one of the most common algorithms in data mining technique. It has the advantage of a high rate of correction and a strong resistance to noisy data. The most widely applied one is BP NN algorithms.Inheriting BP NN algorithms is an inheriting algorithms using tri-chromosome structure to optimize BP NN。This method can optimize BP NN connection weight and network topology at the same time, as well as overcome the shortcomings of using NN algorithms alone.In biology, various genes of individuals interact with each other. In some cases, some genes are active and some are recessive. Moreover, higher level genes control the lower level genes. We simulate the phenomenon in biology and use three layers of chromosome structure to design inheriting BP NN。The first layer is the recessive gene which is coded in binary system. It is used to present the hiding data in BP NN。The length of codes for this layers shows the maximum value of network hiding layer.The second layer is never fiber gene groups, which is coded in binary system. It shows the number of never fiber under a specific hiding layer。The third layer employs decimal system. It shows the threshold and weight of upper level never fiber .However, BP NN algorithms suffers the problem of low efficiency, low speed of convergence and high tendency to get trapped in local minimum. To solve those problems, this paper proposes a method in using inheriting...