基于银行大数据的用户信用风险预测模型

User Credit Risk Prediction Model based on Big Data

摘要: 信用风险是银行经营的主要风险，影响银行的发展，有必要建立信用风险预测模型，帮助银行规避风险、减少损失。本文以某家商业银行的八万条千维数据作为研究对象，采用“分组主成分”的方法对千维变量进行降维的数据预处理，运用Logistic回归和随机森林建立信用风险预测模型。两种模型的分析结果显示，客户的信用卡级别、职业、价值等级、个人业务基本情况、存款及本外币持有额情况对违约风险预测的影响较大。Logistic回归曲线下面积为0.847，预测准确率为75%；随机森林曲线下面积为0.848，预测准确率为85%，相较于以往的研究，两个模型的预测准确率都有明显提高。实际应用时，两种模型可以相互结合，充分发挥二者的优越性。

Abstract: Credit risk is the main risk of bank operation and affects the development of bank. It is necessary to establish credit risk prediction model to help banks avoid risks and reduce losses. In this paper, 80,000 pieces of thousand dimensional data of a commercial bank are taken as the research object, and the method of “group principal component” is used to preprocess the data of thousand dimensional variables. Then, the credit risk prediction model is established by using Logistic regression and random forest respectively. The analysis results of the two models show that the customer’s credit card level, occupation, value level, basic information of personal business, deposits and foreign current holdings have great influence on predicting the probability of default. The area under the curve of logistic regression model is 0.847, and the prediction accuracy is 75%; the area under the curve of the random forest model is 0.848, and the prediction accuracy is 85%. Compared with previous studies, the prediction accuracy of the two models is significantly improved. In practical application, the two models can be combined with each other to give full play to their advantages.