Speaker: Kseniia Kurishchenko, CBS

Title: Fair Heterogeneous Treatment Effect Forests

Abstract: Nowadays, it is common to use observational data in the decision-making process. However, observational data may be biased against a group defined by a sensitive attribute such as gender, race, etc. If not carefully trained, the algorithm may provide discriminatory results. In this talk, I develop a model that evaluates and predicts the heterogeneous treatment effects to decide who is targeted for treatment, while ensuring that on average the sensitive group treatment effect does not differ much from the non-sensitive one. As the starting point, I use a Random Forest to define similar individuals. Each leaf of the Random Forest provides a neighborhood of individuals and a linear model predicting the treatment effect. I introduce a Mathematical Optimization model that combines all the linear models and reweighs them in order to have an accurate heterogeneous treatment effect prediction and a good level of fairness. I present simulated data results illustrating that my model provides fairer predictions of the treatment effect than the benchmark.

Kseniia Kurishchenko is a PhD student at CBS. Her areas of interest are Econometrics, Machine Learning and Operations Research. She has a Bachelor's degree in Applied Mathematics and a Master's degree in Economics.

Host: Guðmundur Stefán Guðmundsson

Organisers: Luke Taylor and Leopoldo Catania.