神经网络技术在P2P网贷中小企业信用评估中的应用

Application of Neural Network Technology in Credit Evaluation of P2P Network Credit SMEs

P2P网络借贷被认为是解决中小企业融资难问题的有效途径,但为规避中小企业的违约风险,中小企业信用评估问题的重要性不言而喻。首先,利用P2P平台--点融网的相关中小企业数据特征以及宏观经济指标,构建27个指标的中小企业信用风险评估指标体系;其次,利用Python技术收集点融网上中小企业的相关数据和宏观经济数据作为训练集和测试集样本,采用SMOTE算法解决训练集中样本的不均衡问题;最后,为达到分类降维的目的,用主成分因子分析方法对信用风险指标进行筛选。然后采用两种神经网络技术--多层感知器与径向基函数--来预测中小企业的违约概率。结果表明,多层感知器对于中小企业违约概率的预测能力较强,准确率达到100.00%,预测非违约中小企业违约风险的准确率仅为76.30%,而径向基函数预测正确率略胜一筹,达到92.50%,总体正确率为93.00%。

P2P network lending is considered to be an effective way to solve the financing difficulties of SMEs.However, in order to avoid the default risk of SMEs, the importance of SME credit assessment is self-evident.Firstly, using the P2P platform-related data characteristics of SMEs and macroeconomic indicators, we will build 27 indicators of SME credit risk assessment index system. Secondly, use Python technology to collect relevant data and macroeconomics of SMEs on the Internet. As the training set and test set samples, the SMOTE algorithm is used to solve the imbalance problem in the training set. Finally, in order to achieve the purpose of classification and dimension reduction, the principal component factor analysis method is used to screen the credit risk indicators. Two neural network techniques-multilayer perceptrons and radial basis functions-are then used to predict the probability of default for SMEs. The results show that the multi-layer perceptron has a strong predictive ability for the default probability of SMEs, and the accuracy rate reaches 100.00%. The accuracy rate of non-default SME default risk is only 76.30%, while the radial basis function prediction accuracy is slightly better, reaching 92.50%, the overall correct rate is 93.00%.