Proposition of an utilitarianism and fair objective function building method based on values and socio-economic consequences for data-driven decisions

Abstract

In this short article, we propose a method to setup the objective function of machine learning binary classifier used in data driven decision. The goal is to take fair decisions aligned with an ethical value system and based on the long-term consequences of prediction errors for all stakeholders.The proposed method is based on human in the loop with an ethical committee to define the appropriate setup of the objective function, depending on the context of the decision. The setup parameters are of three categories: the fairness criteria, the ethical values and the weights associated to socio-economic long-term consequences of prediction errors.

Keywords: Data-driven decision fairness utilitarianism values

DOI: 10.54941/ahfe1004082