Nearest Neighbor Clustering Method And Rbf Neural Network-based Commercial Bank Branches Performance Evaluation Studies

With China's accession to WTO,competition among banks is much fiercer.From December 11 in 2006,China's financial gateway aready have opened to the world financial institutions.Foreign banks has been gradually come into China which increasing competition in the banking sector. The global financial tsunami which swept all the countries has worsen the banking industry over the recent years.Therefore, it is urgent to improve the management of China's commercial banks to deal with these crises,and meanwhile ,we should pay more attention to performance evaluation of operating results of commercial banks.It is a serious problem placed in front of commercial banks that using scientific methods to evaluate the performance of commercial banks comprehensively. At presnet ,most of the studies of the relevant subjects are the head offices'performance of commercial bank ,and branches of commercial banks operating performance are fewer.So,building a comprehensive and accurate branch performance evaluation system and using which kind of evaluation methods to reflect the performance evaluation of a bank branch performance is one of essencial steps .In this paper ,on the basis of the results in the performance evaluation of commercial banks at home and abroad, the concepts and theories of performance evaluation of commercial banks as a starting point, analysing of the existing branches of Commercial Bank Performance Evaluation System defects, using the Delphi method to conduct a preliminary selection of indicators,then adjust and improve the indicators by mathematical statistics and qualitative analysis method of combining,Establishing a three-level index of 27 branches of commercial banks performance evaluation index system.In the past methods of branches of commercial banks performance evaluation,traditional-fuzzy comprehensive evaluation method can not get rid of the randomness of decision-making process,subjective uncertainty and lack of pacing with times the shortcomings of self-learning ability。And in the past, BP neural network method of evaluating had the shortcomings of local extremum. Although it owned self-learning ability,however, self-learning training it has fatal flaws: lacking of rigorous experimental design in the case of limited learning samples, thus, this will lead that the trained network is difficult to cover most cases and neural network can not be trained, and have poor evaluation of performance.What'more ,in this paper,I use the uniform design method to design RBF neural network training samples.The uniform design method,With fewer number of tests to arrange multi-factor, multi-level factorial experiment,is the best measure of uniformity under the best method of factorial experimental design. It will generate uniformly distributed, large, and representative samples to train RBF neural network.We exert uniform design method of U1000 (527) ,to obtain large and standard uniform design samples by DSP Software,and then apply RBF neural network training on these representative samples. We get the results which its error is less than the scheduled error of 0.0001, and obtain instance test results which its error is less than scheduled instance error of 0.2. So we establish a set of accurate, fast, intelligent structure of the bank branch performance evaluation model which enable to estimate our commercial bank branch performance evaluation scientifically and accurately.