摘要
互联网金融是对传统金融模式的延伸,但由于部分借款人在借款后无法按期、足额还款,使得互联网金融平台面临着信用风险。对借款人的信用风险进行准确评估,可以降低风险,并且能够在一定程度上为互联网金融行业的稳定发展提供保障。数据分析方法在信用风险评估领域已有广泛应用。本文从国内某互联网金融平台借款人的个人、资产、借款信息三类数据提取特征,研究了数据分析方法中Logistic回归的衍生方法逐步Logistic回归、弹性网络和Bagging集成方法的代表Bagging、极端随机树和随机森林。研究发现随机森林与逐步Logistic回归分别在F1-score、Accuracy、FPR和AUC指标下效果最优,且筛选出的重要特征也保持一致。
Internet finance is an extension of the traditional financial model. As some borrowers are unable to repay in full or on time, the platform faces credit risks. Accurate assessment of the credit risk can reduce losses and provide a guarantee for the stable development of the Internet finance industry. Data analysis methods have been widely used in the field. This paper extracted features from the three types of personal, asset and loan information. Stepwise Logistic regression and Elastic Net which are the derivative methods of Logistic regression, and the representative of Bagging ensemble method, Bagging, Extremely Randomized Trees and Random Forest were studied. It is found that Random Forest and Stepwise Logistic regression have the best results under F1-score, Accuracy, FPR and AUC indicators respectively, and the important features selected are also consistent.
出处
《应用数学进展》
2022年第4期1657-1667,共11页
Advances in Applied Mathematics
