基于Wrapper的信用风险预测模型研究与应用

Research and application of credit risk prediction model based on wrapper

研究了一个有效适用的企业信用风险预警模型。针对单一BP神经网络预测模型由于财务指标选择不当导致误判率较高的问题,提出了首先进行特征选择,利用遗传算法搜索出最优特征子集,并采用BP神经网络作为遗传算法的评估函数,构建了一个基于Wrapper方法的神经网络信用风险预测模型。以沪深股市1998—2004年间的制造企业数据为例对模型进行实验,结果表明,新模型提高了预测准确率,评估结果更具科学性,实际应用具有良好的信用风险预测能力。

The paper studies an effective enterprise credit risk prediction model based on the wrapper methods. For single BP neural network prediction model with higher rate of misjudgment due to improper financial index selection, the paper puts forward a BP neural network learning algorithm, which first uses the conclusion of feature selection, then gets optimal feature subset, and uses the BP neural network as the evaluate function of genetic algorithm, constructed a Wrapper method based on neural network to credit risk prediction model. In the paper, an example is adopted to test the new model by using the data of Hu-Shen stock market from 1998 to 2004 in the manufacturing enterprise. The simulation results show that the new model improves the predic- tion accuracy, evaluate the results more scientific and practical application with good credit risk.