fairadapt: Causal Reasoning for Fair Data Pre-processing

Abstract

Machine learning algorithms are useful for various predictions tasks, but they can also learn how to discriminate, based on gender, race or other sensitive attributes. This realization gave rise to the field of fair machine learning, which aims to measure and mitigate such algorithmic bias. This manuscript describes the R-package fairadapt, which implements a causal inference pre-processing method. By making use of a causal graphical model and the observed data, the method can be used to address hypothetical questions of the form "What would my salary have been, had I been of a different gender/race?". Such individual level counterfactual reasoning can help eliminate discrimination and help justify fair decisions. We also discuss appropriate relaxations which assume certain causal pathways from the sensitive attribute to the outcome are not discriminatory.

Network-based clustering unveils interconnected landscapes of genomic and clinical features across myeloid malignancies

“…These datasets contain sensitive attributes such as gender and race, which were utilised to measure the fairness of the clustering algorithms. The pre-processed UCI Adult and COMPAS datasets were downloaded from the fairadapt package [Plečko et al, 2021] and the pre-processing procedure is described in detail in [Plečko and Meinshausen, 2020]. The datasets were pre-processed by excluding features such as relationship, final weight, education, capital gain, and capital loss.…”

Reconciling Predictive and Statistical Parity: A Causal Approach

“…If one is also interested in constructing a new fair predictor before using Alg. 1 (instead of testing an existing one), one may use tools for causally removing discrimination, such as (Chiappa 2019) or (Plečko and Meinshausen 2020;Plečko, Bennett, and Meinshausen 2021). In Appendix E we show formally that a predictor Y satisfying the conditions of Alg.…”