基于动态聚类RBF网络的小企业信贷预测研究

ON SMALL BUSINESS CREDIT FORECASTING BASED ON DYNAMIC CLUSTERING RBF NETWORK

利用动态聚类算法确定RBF网络的隐含层节点,不仅聚类速度快,而且隐含层节点数的优化提高了网络的利用效率;定义了广义重要度欧氏距离用于算法中的距离计算;根据穆迪、安德尔违约概率曲线定义了信用评级风险系数等指标。最后,以南京某商业银行数据为依据,利用Matlab为工具平台,建立基于动态聚类的RBF神经网络模型。实证分析表明:本信贷预测模型对违约小企业的判别准确率较高,可为银行有效地甄别高风险企业。

In this paper,we use dynamic clustering algorithm to determine hidden layer nodes in RBF neural network,the clustering speed is fast,and the optimization of the number of hidden layer nodes improves the utilisation efficiency of the network as well.The general important degree Euclidean distance is defined to calculate the distance in the algorithm;According to Moody,Alder＇s default probability curve,we define the indices such as the risk coefficient of credit rating.At last,we set up a dynamic clustering algorithm-based RBF neural network model based on the data of a certain commercial bank in Nanjing by using Matlab as the tool platform.Demonstration analysis shows that the credit forecasting model has higher distinguishing accuracy on defaulting small business;and is able to discriminate high-risk enterprises for the banking.