带权重的RBF神经网络银行个人信用评价方法

Method of personal credit evaluation of bank based on RBF neural network with weight

针对RBF神经网络确定核函数中心时没有考虑输入样本分类指标权重的问题,提出了一种动态加权聚类算法。在算法中利用样本之间的加权距离代替了欧氏距离作为选定核函数中心的量度。在此基础上,建立了信用评价模型,利用已知类别的样本对模型进行训练,再利用训练好的模型对未知类别的样本进行预测,实验结果验证了模型的有效性。

A dynamic weighting cluster algorithm is proposed in this article in view of the problem of input sample＇s classifica- tion weight being not considered by formerly RBF neural network. In this algorithm, the weighting distance replaces the Euclidean distance to act the role of measurement to the cluster. Based on this, the credit evaluation model is established, which is trained by known category sample. Then the trained model is used to forecast the unknown category sample, the experimental result confirms the model＇s validity.