基于LSTM-CNN的中小企业信用风险预测

Credit Risk Prediction of Small and Medium-Sized Enterprises Based on LSTM-CNN

针对中小企业的信用风险预测问题,文章提出了一种基于长短期记忆网络(Long Short-Term Memory,LSTM)-卷积神经网络(Convolutional Neural Network,CNN)的中小企业信用风险预测方法.首先,依据国标《企业信用评价指标》,结合中小企业的特点,构建中小企业信用风险预测指标体系,包括守信意愿、守信能力和守信表现三方面的财务与非财务指标;然后,优化LSTM-CNN的网络结构和参数,并使用Dropout方法与Batch归一化方法防止过拟合;最后,采集上市中小企业数据,对数据进行缺失值、标准化与过采样处理,利用LSTM-CNN自动提取信用风险特征,并进行信用风险预测.实验结果表明,文章构建的指标体系能够更为全面的反映信用风险状况,基于LSTM-CNN的中小企业信用风险预测效果优于其他信用风险预测模型,克服了传统方法无法对中小企业时序数据进行动态预测、忽视中小企业发展潜力与时间延续性的局限.

Aiming at the credit risk prediction of small and medium-sized enterprises,this paper proposes a credit risk prediction method based on Long Short-Term Memory(LSTM)-Convolutional Neural Network(CNN)of small and medium-sized enterprises.Firstly,according to the national standard"Enterprise Credit Evaluation Index"and the characteristics of small and medium-sized enterprises,this paper proposes a credit risk prediction index system of small and medium-sized enterprises.The index system includes three kinds of financial and non-financial indicators:Credit intention,credit ability and credit performance.Then,this paper optimizes the network structure and parameters of LSTM-CNN,and applies Dropout and Batch Normalization methods to prevent over fitting.Finally,collecting the information of the listed small and medium-sized enterprises,and after missing value processing,standardization and oversampling,LSTM-CNN is applied to automatically extract features and predict credit risk.The experimental results show that the index system of this paper comprehensively reflect the credit risk situation.The credit risk prediction effect of small and medium-sized enterprises based on LSTM-CNN is better than the comparative models,which overcomes the limitations of traditional methods that cannot dynamically predict the time series data,and ignore the development potential and time continuity of small and medium-sized enterprises.