智能科学与技术学报 ›› 2019, Vol. 1 ›› Issue (4): 342-351.doi: 10.11959/j.issn.2096-6652.201939

• 学术论文 • 上一篇    下一篇

数字金融时代机器学习模型在实时反欺诈中的应用与实践

曹汉平,张晓晶,祝睿杰(),黄潇拉   

  1. 中国银行总行个人数字金融部,北京 100818
  • 修回日期:2019-11-12 出版日期:2019-12-20 发布日期:2020-02-29
  • 作者简介:曹汉平(1976- ),男,博士,中国银行总行个人数字金融部副总经理,主要研究方向为宏观经济、大数据系统架构与应用、新型渠道、数字金融风险防控等|张晓晶(1980- ),女,中国银行总行个人数字金融部高级经理,主要研究方向为机器学习建模、智能风控系统建设、数字金融业务风险防控等|祝睿杰(1986- ),男,中国银行总行个人数字金融部业务经理,主要研究方向为数据挖掘、机器学习建模、数字金融业务风险防控等|黄潇拉(1988- ),女,中国银行总行个人数字金融部业务经理,主要业务方向为数据分析、智能风控系统建设、数字金融风险防控等

Application and practice of machine learning model in real-time anti-fraud in the era of digital finance

Hanping CAO,Xiaojing ZHANG,Ruijie ZHU(),Xiaola HUANG   

  1. Digital Personal Banking Department,Bank of China,Beijing 100818,China
  • Revised:2019-11-12 Online:2019-12-20 Published:2020-02-29

摘要:

近年来,数字金融蓬勃发展,金融科技日趋成熟,信息技术的发展对社会产生巨大积极作用的同时也带来了新型风险,网络黑产呈爆发式增长,电信网络诈骗给人民群众造成了巨大的财产损失。在数字金融时代,商业银行既迎来了新的机遇与动力,又面临着新的挑战和数字化变革的要求,线上金融业务已经成为新的主战场。在此背景下,基于 RFM 高维衍生特征和对机器学习算法的研究,构建了基于高维交易行为画像的事中反欺诈机器学习模型。依托大数据、流计算等技术,通过在系统化部署、应用策略以及模型迭代优化等环节的实践,形成了一套基于机器学习模型的事中风控方案。实践证实,该模型的AUC达到了0.972,可以敏锐洞察欺诈风险,实现毫秒级的欺诈交易识别,对于提升线上数字金融业务的事中风控能力具有一定的推广价值和借鉴意义。

关键词: 数字金融, 机器学习, 反欺诈, RFM, 流计算

Abstract:

In recent years,with the rapid development of FinTech,digital finance has flourished and brought huge positive effect on society.Meanwhile,new risks have been introduced into banks.For example,the black production related to network security has experienced explosive growth,and telecommunication network fraud has caused property losses to the public.In the era of digital finance,the commercial banks have not only ushered in new opportunities and dynamics,but also faced new challenges and requirements for digital transformation.As a result,e-finance has become a new battlefield.With this context,a real-time anti-fraud machine learning model based on high-dimensional transaction behavior portrait through enhanced RFM feature-derivation and machine learning modeling was established in this paper.Relying on the new technologies such as big data,stream computing,a model application solution to real-time risk control was formed including systematic deployment,model application strategies and iterative model optimization.Through practical observation,the AUC of the model reaches 0.972,which provides a keen insight into fraud risk,realizes millisecond-level risk identification,and promotes risk control ability of e-finance significantly.

Key words: digital finance, machine learning, anti-fraud, RFM, stream computing

中图分类号: 

No Suggested Reading articles found!