知识卡片

普通读书笔记卡 · 6564.a · 生成模型:GPT-6

拦截交易造成选择性标签

理解等级:中级

欺诈系统拒绝交易后,可能无法观察其若放行是否会被报告为欺诈。新训练数据只保留通过旧策略的交易,标签缺失因此与旧预测相关。仅在已放行样本上再训练,可能掩盖被拦截人群的错误;书中未给出具体纠偏估计方案。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      customers. Those transactions are labeled fraudulent in this new block of data, but the DGP has changed from the initial training set. The deployed version 1.0 predictive model is rejecting transactions that it predicted were fraudulent, but because they were rejected, you don’t know if they were actual cases of fraud. These rejected transactions are excluded from the next training set because they lack labels.

    参考来源

    - 位置:《Causal AI》第07章《干预与因果效应》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Causal AI/_epub-src/OEBPS/Text/chapter-7.xhtml) - 结论依据:摘录中关于拦截交易造成选择性标签的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:customers. Those transactions are labeled fraudulent in this new block of data, but the DGP has changed from the initial training set. The deployed version 1.0 predictive model is rejecting transactions that it predicted were fraudulent, but because they were rejected, you don’t know if they were actual cases of fraud. These rejected transactions are excluded from the next training set because they lack labels.