On Fair Lasso Regression

In this paper we introduce an approach to incorporate fairness constraints into LASSO regression. Assuming that a group of individuals need to be protected against discrimination, we address the problem of training the LASSO regression subject to a fairness constraint, enforcing that an equal proportion of individuals from the protected and non-protected groups have a predicted value above a given threshold. The LASSO model is then extended to account for this and is formulated as a Mixed-Integer Quadratic Program with linear constraints. We illustrate on real-world datasets that we are able to significantly improve fairness, in terms of our novel unfairness measure, without incurring a significant loss in prediction accuracy or sparsity.