On discrimination discovery using causal networks

Abstract

Discrimination discovery is an increasingly important task in the data mining field. The purpose of discrimination discovery is to unveil discriminatory practices on the protective attribute (e.g., gender) by analyzing the dataset of historical decision records. Different types of discrimination have been proposed in the literature. We aim to develop a framework that is able to deal with all types of discrimination. We make use of the causal networks, which effectively captures the existence of discrimination patterns and can provide quantitative evidence of discrimination in decision making. In this paper, we first propose a categorization for various discrimination. Then, we present our preliminary results on four types of discrimination, namely system-level direct discrimination, the system-level indirect discrimination, group-level discrimination, and individual level discrimination.We have conducted empirical assessments on real datasets. The results show great efficacy of our approach.