Casual Inference

[btanasie]

Looking to use causal inference to improve the understanding of our model and data.

Some potential ideas:

• Controle for features and their impact of categorizing descriptions
• Look at demographic data and influence factors to explore data bias

DoWhy would the main tool for this

[Andrea-Yzeiri]

As mentioned in class, because of the nature of our predictive model, predicting on description (multi-class categorical), Uplift tree would be better.

[btanasie]

Yes agreed! Ended going with a new package called CausalLift (https://github.com/Minyus/causallift) to perform this for each demographic variable as treatment.