中文摘要：

广义回归神经网络GRNN和概率神经网络PNN，与传统的BP神经网络相比，收敛速度快，学习能力强．本文将其应用到信用风险评估，选取1057组公司财务数据作为训练数据，350组数据作为测试数据，分别建立基于不同属性的模型对样本公司财务状况评判其是守信公司还是违约公司，最终选取精度较高的作为最终模型对财务系统进行预测．结果表明，PNN对于信用风险评估泛化能力好，测试集正确率高，因此可以用作风险预警的模型，给决策者提供智力支持．

英文摘要：

Since the generalized regression neural network （GRNN） model and the probabilistic neural network （PNN） model possess better convergence rate and learning ability than the traditional BP neural network model, we apply them to the selected companies financial data to assess their credit risk, where 1057 groups of data are used as training data, and the rest 350 groups of data are used as test data. Models of different attributes are established based on the sample companies＇ financial situation for judging whether a company is defaulting： Models with high precision are selected to forecast the financial system. Empirical results show that the PNN model has good perform- ance and high accuracy for credit risk assessment, and thus can be used for risk alert and providing intellectual support for decision makers.