Counterfactual Fairness in Synthetic Data Generation

Mahed Abroshan, Mohammad Mahdi Khalili, Andrew Elliott

Abstract: Synthetic data generation (SDG) is proposed as a promising solution for data sharing as in many high-stake applications due to privacy concerns, releasing the real dataset is not an option. While the main goal of private SDG is to create a dataset that preserves the privacy of individuals contributing to the dataset, the use of synthetic data also creates an opportunity to improve the fairness issue at the source. Since there exist historical biases in the datasets, using the biased data to train an ML model can lead to an unfair model which may exacerbate the discrimination. Using synthetic data, we can attempt to remove the bias from the dataset before releasing the data. In this work, we formalize the definition of fairness in synthetic data generation and propose a method to achieve counterfactual fairness.