基于随机森林的第三方支付违规风险预警研究

Research on the Third‑party Payment Violation Risk Early‑warning Based on Random Forest

立足于第三方支付行业发展中存在的突出问题,本文给出基于随机森林的违规风险预警机制并讨论具体实施。首先,构建风险预警指标体系,结合机器学习中的随机森林算法,提出风险预警机制。然后,以已获支付牌照的271家企业为样本,验证所提出的违规风险预警机制的有效性。通过对比随机森林模型和Logistic模型的判定结果,发现随机森林显著降低了一类错误率和二类错误率,模型正确预测率高达99.01%。最后,通过对指标体系中的重要变量进行分析,提出具体应用措施及相应的风险监管建议。

For the various violation phenomena existing in the third‐party payment industry,a violation risk early-warning mechanism based on Random Forest are proposed.A set of index system for violation risk early-warning is firstly built,then the risk early-warning mechanism based on random forest algorithm which comes from machine learning methods is proposed.A sample of 271 companies with paid licenses is used to verify the effectiveness of the proposed early warning mechanism for violation risks.By comparing the discrimination rate of Logistic model and Random Forest model,the research finds that both type Ⅰ error and type Ⅱ error of Random Forest model are much lower,and Random Forest model has high accuracy of up to 99.01%when predicting.Finally,through the analysis of important variables,specific application measures and corresponding risk supervision suggestions are put forward.