Research On Improved General Fuzzy Min-Max Neural Network And It’s Application

In recent years, fuzzy theory and neural network technology in the field of artificial intelligence research is the two hot spots. And then we combine them organically and form a fuzzy neural network (FNN). FNN is a neural network which the input signal is fuzzy variable. FNN absorbs the advantages of neural network and fuzzy system, it has superiority in many problems, especially in challenge of nonlinear, fuzzy. There is a huge potential in terms of intelligent information processing. On the basis of fuzzy neural network, general fuzzy min-max neural network(GFMN) is presented. This network has distinctive features, which is a fusion of clustering and classification algorithm.This paper has two main parts. The first section makes a systematic summary to GFMN. A new algorithm is presented for improved the and implemented in the second section.GFMN is capable of labeling data and unlabeling data simultaneously. In pattern recognition, learning the part labeling data is important. Of course, how to study identification data effectively is also essential. But the network has some defects. It can’t completely clustering and adaptive online learning, training data still needs to be part identification. When faced with a new class, it will be divided into the unknown class, and then can’t reach clustering the desired effect.In this paper, the improved GFMN fills this shortage. In the process of learning the experimental samples, the network tries to accommodate the sample in one of the existing hyperboxes of that class, the hyperboxes size will be changed, so this process is known as hyperbox expansion. If the hyperbox expansion does not exist, a new hyperbox is added to the network, and belonging to the same class of hyperboxes will form a complete class. By modifying its structure and learning algorithm, this paper presents a adding new class or deleting an existing class of GFMN and solves the shortcomings existing in the original algorithm. For the application of GFMN, analysis of enterprise credit evaluation as an example to verify the proposed network structure and learning algorithm is effective.