RBF Neural Network Ensemble Research And Its Application To Personal Credit Evaluation

The technology of neural network ensemble is a neural computing research focus, which already have mature application in many areas. Neural network ensemble is a learning paradigm where many neural networks are jointly used to solve a problem. Neural network ensemble has become a very successful technology. The neural network ensemble method mainly concentrates in two areas, namely, how to train a number of individual neural networks and how to combine the individual predictions. Theory and practice have proved that, by improving the accuracy of the individual neural network and the differences between neural networks, the generalization of neural networks ensemble can be enhanced.Evolutionary Programming (EP) has the following characteristics: (1)EP based on Lamarckian evolution is put emphasis on the behavioral link between parents and their offspring;(2)In order to reduce the detrimental effect of the permutation problem, EP algorithm, which use mutation, rather than crossover, is adopted; (3)Node deletions are always attempted before node additions in the mutations in order to encourage the evolution of small RBFNNs; These features of evolutionary programming(EP) make it possible to obtain big difference individual networks through the use of EP to evolve neural networks.Cultural algorithm is a new global optimization algorithm used to solve complex calculation. The algorithm simulates the evolutionary processes of human society. In human society, culture can be seen as the carrier of information, such information potentially affects all members of society, and is benefit to guide the same generation and the future generations of problem-solving practice. Similarly, the important idea of culture algorithm lies in the acquisition of knowledge (that is, beliefs) of problem to be resolved from the evolutionary population and feedback of this knowledge to guide the search process.On the base of analysis of neural network ensemble methods,evolutionary programming, and cultural algorithm,this paper presents a more effective neural network ensemble method, and applies it to personal credit evaluation. The main work of this paper appears in the follow aspects:1. On the base of research on the training of individual neural networks, this paper presents a neural network ensemble method––neural network ensemble method based on evolutionary programming(EP-ANNE). In the further training of individual networks, we introduce the evolutionary programming method. The specific implementation is as follows: the hidden layer neural network centers use real matrix to describe, and we have introduced a negative correlation learning mechanism to choose the fitness function.We use rank-selection strategy to choose individual networks to train,and the results of individual networks training, that is, "success" or "failure" , determine whether or not to have structural variation, thereby bringing about the network topology negative correlation training. Through introducing the evolutionary programming to build the ensemble,we can get heterogeneous RBF networks with big difference, so the generalization of ensemble is enhanced. 2. On the base of research on the choice of individual neural networks for ensemble, this paper presents a neural network ensemble method––neural network ensemble method based on the cultural algorithm(CA-ANNE).This paper presents an approach for neural network ensemble based on culture algorithm, in which the culture algorithm is used to select part of the trained individual networks to be ensembled. This method puts genetic algorithm into the framework of culture algorithm which can fully utilize the outstanding characteristics of the individual that greatly improves the speed of the search algorithm. The experimental results proved that culture algorithm is can increase the structural difference between individual neural network and can effectively improve the efficiency to select diversity individual neural networks to construct ensemble.3. Finally, the proposed neural network ensemble methods are applied to personal credit evaluation. By comparing the experimental results from different perspectives, we prove the effectiveness of the methods.