组合模型在商业银行信用风险评估中的研究

Modeling Based on RS and BPNN for Credit Risk Assessment in Commercial Banks

研究商业银行信用风险评估问题,商业银行信用风险评评估涉及指标相当多,各指标间呈非线性关系且存在严重冗余信息,传统评估方法不能很好消除冗余信息,只能反映指标间的线性关系,导致风险评估准确率低。为了提高商业银行信用风险评估的准确性,提出了一种粗糙集理论(RS)和BP神经网络(BPNN)相结合的商业银行信用风险评估组合模型(RS_BPNN)。新模型首先利用粗糙集理论对各评估指标进行指标约简,消除指标间的冗余消息,简化神经网络的网络结构,然后将约简后的数据输入非线性预测能力优异的BP神经网络进行训练,得到商业银行信用风险评估模型,最后采用中国工商银行某分行数据对组合模型进行仿真试验。仿真结果表明,与传统的BP神经网络模型相比,组合模型加快了网络的运算速度,提高风险评估准确率,获得评估结果更具科学性。

Study commercial bank credit risk assessment. Assessing credit risk in commercial banks, involves many appraisal targets, and the indexes have much redundant information, therefore, the traditional method is unable to eliminate these redundancies, and the result precision is not very high. ~In order to improve the credit risk assess- ment of commercial banks, a combined commercial bank credit risk assessment model （RS_BPNN） of rough set theo- ry （RS） and the BP neural network （BPNN） is put forward. Firstly, rough sets theory of numerical analysis of strong capability evaluation indexes is used for attribute reduction, then the BP neural network training data are reduced to simplify the network structure. Secondly, the reduced BP neural network is trained. Finally, the simulation experi- ment is carried out. The results show that compared with the traditional BP neural network model, the combined mod- el speeds up the network operating speed, further enhances assessubg precision, and has obtained the good appraisal result.