机器学习-算法理论

Confounding Bias & Selection Bia

2021-06-30  本文已影响0人  shudaxu

Confounding bias [1]& Selection bias[2]的差异与相似处。

之前的文章从直觉上讨论了系统中的各种bias。
其实Selection Bias这个问题并不仅仅是在Causal Inference中有讨论,在单纯的Prediction任务中也有讨论。

解决办法

Refer
[1]Propensity score
https://blog.csdn.net/a358463121/article/details/106911783

[2]本篇讨论什么样的Selection Bias可以恢复(1、恢复概率分布,2、恢复Causal Effect Estimation)注意,这里是Selection Bias,不是Confounding
Recovering from Selection Bias in Causal and Statistical Inference

[3] Identifying Confounding?可行性?causal discovery从数据角度的无解性如何identify?所以,我们可以验证它在数据上的存在性,但是不能确切地从数据直接发现confounding。
https://sphweb.bumc.bu.edu/otlt/mph-modules/bs/bs704-ep713_confounding-em/bs704-ep713_confounding-em_print.html

[4]CER中的研究(Clinical Medicine,临床医学中常用的实验,比如判断Covid疫苗的有效性)
Distinguishing selection bias and confounding bias in comparative effectiveness research

[5]
Linear Models: A Useful “Microscope”
for Causal Analysis

[6]Censoring 章节 in Survival_analysis
https://en.wikipedia.org/wiki/Survival_analysis

[7]Sampling Bias
https://en.wikipedia.org/wiki/Selection_bias

[8]譬如:
Improving Ad Click Prediction by Considering Non-displayed Events
以及https://zhuanlan.zhihu.com/p/342905546

上一篇下一篇

猜你喜欢

热点阅读