博客笔记五: [Airbnb] 信用卡欺诈的loss function设计

原文标题:Fighting Financial Fraud with Targeted Friction

作者:David Press

原文地址: https://medium.com/airbnb-engineering/fighting-financial-fraud-with-targeted-friction-82d950d8900e

面对问题:

别人使用偷来的卡,真正持卡人退款

目标:

false positives: “good” events that a model or rule classifies as “bad” (above the threshold,positive (fraud) and negative (non-fraud) ),提高threshold,宁可错杀一千,不可放过一个。因而对于大多数被误解的用户,增加了付款难度(additional verification called a friction)。

难度增加包括:网银登录,三位码,地址信息(比如邮编)

To stop the use of stolen credit cards, our chargeback model triggers a number of frictions to ensure that the guest is in fact authorized to use that card, including micro-authorization (placing two small authorizations on the credit card, which the cardholder must identify by logging into their online banking statement), 3-D Secure (which allows credit card companies to directly authenticate cardholders via a password or SMS challenge), and billing-statement verification (requiring the cardholder to upload a copy of the billing statement associated with the card).
难点: imbalanced classification

loss function :

对于不同的错误给予不同的weight,但是false positive漏过欺诈的代价最大
1. false positive:
G: 因为friction好人离开率
V: lifetime value好人价值
每一个friction影响都有ab test测试,但是测试太贵了,需要highly imbalanced test(缺点是相比于50/50, 需要太久收敛)
2. false negative:
没有检测出来的漏网之鱼,只能损失照单全收
3. true positive:
判断正确,但是不一定friction可以阻挡,F 需要新的ABtest来决定。F=1说明friction 100%阻挡了欺诈操作
总结: 两个ab test示意图

最终模型效果

模型判断不同threshold带来不同的总loss,因为好人离开损失和吸收坏人的损失是相反的。

总结:
- 介绍信用卡欺诈判断模型loss function的细节,更加清晰了
- 介绍了典型的两个abtest应用场景,很有帮助
- 没有提到模型,连一嘴都没有,有点失望

你可能感兴趣的:(机器学习,读博客笔记,官方博客笔记)