| 基于贷款人视角的互联网金融信用风险分级研究 |
| |
| 引用本文： | 林奕皓,王宇森,李旭东,许永峰.基于贷款人视角的互联网金融信用风险分级研究[J].教育技术导刊,2020,19(6):29-34. |
| |
| 作者姓名： | 林奕皓 王宇森 李旭东 许永峰 |
| |
| 作者单位： | 西北大学 数学学院，陕西 西安 710127 |
| |
| 基金项目： | 国家级大学生创新创业训练计划项目（201910697048） |
| |
| 摘 要： | 为提升互联网金融行业贷款人决策的直观性与层次性，提出一种信用分级模型。对历史样本的信用评价指标进行主成分分析，提取关键信息。利用 Logit 回归模型得到“是否违约”和“评价指标主成分”的关系，依据回归方程所得的“违约概率”对借款人进行信用分级。采用遗传模拟退火算法（GSAA）改进的 BP 神经网络，学习“等级”和“评价指标”间的映射规则。利用 Kaggle 网站信用数据集进行实验，结果表明，Logit 回归结果可信度高，“依概率分级”区分度高，GSAA 算法可有效提升 BP 神经网络的精准分级率。分级模型在测试样本上的可信度为 99.02%，优于二值分类和指标赋权模型，可有效降低贷款人资金风险，推动互联网金融行业高质量发展。
|
| 关 键 词： | 互联网金融 信用等级 Logit 模型 BP 神经网络 遗传模拟退火算法 |
| 收稿时间： | 2020-02-23 |
Credit Grading of Internet Finance Based on Lender’s Perspective |
| |
| LIN Yi-hao,WANG Yu-sen,LI Xu-dong,XU Yong-feng.Credit Grading of Internet Finance Based on Lender’s Perspective[J].Introduction of Educational Technology,2020,19(6):29-34. |
| |
| Authors: | LIN Yi-hao WANG Yu-sen LI Xu-dong XU Yong-feng |
| |
| Institution: | School of Mathematics，Northwest University，Xi’an 710127，China |
| |
| Abstract: | In order to improve the intuitiveness and hierarchy of the lender’s decision in the Internet finance industry，a credit grading model is constructed. Principal component analysis is carried out on the credit evaluation index of historical samples to extract the key information. Logit regression model is used to obtain the mapping relationship between“default or not”and“principal component of evaluation index”. Borrowers are graded by default probability based on the regression equation. By training the improved BP neural network with genetic simulated annealing algorithm（GSAA），the mapping rules between“grade”and“evaluation index”are learned. Samples of borrowers on Kaggle’s site are used for experimental research. The result shows that Logit regression is highly au? thentic and probability grading is highly differentiated. The GSAA algorithm effectively increases the accurate grading rate of BP neural network. The grading credibility of the grading model to the test samples reaches 99.02%，which is higher than the model based on binary classification or index weighting. The grading model can effectively reduce the fund risk of lenders and promote the high-quality development of Internet finance industry. |
| |
| Keywords: | Internet finance credit grade Logit model BP neural network genetic simulated annealing algorithm |
|
| 点击此处可从《教育技术导刊》浏览原始摘要信息 |
| 点击此处可从《教育技术导刊》下载免费的PDF全文 |
