基于Logistic模型的小微贷不良用户画像
Portrait of Non-Performing Users in Small and Micro Loans Based on Logistic Model

随着互联网金融的发展及个人消费需求的日益增长，小微贷，特别是基于互联网的P2P借贷得到了较快的发展。但由于平台风险识别能力的缺失，部分平台产生大量违约情况，致使投资人遭受损失。为辅助平台及投资人有效识别不良用户，减少坏账带来的损失，本文基于国内外较有代表性的P2P平台：Prosper及拍拍贷上的数据，采用逻辑回归、决策树及支撑向量机三类模型对借款人进行信用评估，并依据模型结果得到小微贷不良用户画像。结果表明，逻辑回归模型时间复杂度低，具有优越的可解释性，更加适用于违约因素的研究。并且不良用户借款通常具有高利率、长期限的特点；同时用户本身没有稳定工作，收入较低。而常被我们关注到的性别、年龄以及学历反而影响较低。
With the development of Internet finance and the increasing demand for personal consumption, small and micro loans, especially P2P lending based on the Internet, have developed rapidly. How-ever, due to the lack of platform risk identification ability, some platforms have experienced a large number of default situations, resulting in losses for investors. To assist the platform and investors in effectively identifying non-performing users and reducing losses caused by bad debts, based on da-ta from representative domestic and foreign P2P platforms such as Prospector and PaiPaiDai, this article uses three types of models, namely logistic regression, decision tree, and support vector machine to evaluate the credit of borrowers. Based on the model results, a portrait of non-performing users of small and micro loans is obtained. The results indicate that the logistic re-gression model has low time complexity and superior interpretability, making it more suitable for studying default factors. Non-performing user loans usually have the characteristics of high-interest rates and long-term limits; at the same time, users themselves do not have stable jobs and have lower incomes. The gender, age, and educational background, which we often pay attention to, have a lower impact.