Information Theoretic Measures for Fairness-aware Feature Selection

Abstract

Machine learning algorithms are increasingly used for consequential decision making regarding individuals based on their relevant features. Features that are relevant for accurate decisions may however lead to either explicit or implicit forms of discrimination against unprivileged groups, such as those of certain race or gender. This happens due to existing biases in the training data, which are often replicated or even exacerbated by the learning algorithm. Identifying and measuring these biases at the data level is a challenging problem due to the interdependence among the features, and the decision outcome. In this work, we develop a framework for fairness-aware feature selection which takes into account the correlation among the features and the decision outcome, and is based on information theoretic measures for the accuracy and discriminatory impacts of features. In particular, we first propose information theoretic measures which quantify the impact of different subsets of features on the accuracy and discrimination of the decision outcomes. We then deduce the marginal impact of each feature using Shapley value function; a solution concept in cooperative game theory used to estimate marginal contributions of players in a coalitional game. Finally, we design a fairness utility score for each feature (for feature selection) which quantifies how this feature influences accurate as well as nondiscriminatory decisions. Our framework depends on the joint statistics of the data rather than a particular classifier design. We examine our proposed framework on real and synthetic data to evaluate its performance.

A Review of Partial Information Decomposition in Algorithmic Fairness and Explainability

“…In Section 4, we review how PID can help in assessing the contributions of either features or data points with applications in feature selection (as discussed in [31]). Related works include [35][36][37][38]. In Section 5, we discuss another avenue where PID plays an important role: quantifying tradeoffs between different measures, as we illustrate through the example of local and global fairness in federated learning (as discussed in [34]).…”

“…A closely related direction of research that bridges fairness and explainability is the problem of feature selection for algorithmic fairness [35][36][37]53]. In [35,37], the authors propose novel information-theoretic techniques that leverage conditional mutual information with the goal of selecting a subset of features that would achieve fairness, in particular, justifiable fairness [47].…”

“…In [35,37], the authors propose novel information-theoretic techniques that leverage conditional mutual information with the goal of selecting a subset of features that would achieve fairness, in particular, justifiable fairness [47]. In [36], the authors explore the problem of feature selection for algorithmic fairness and propose techniques that leverage partial information decomposition to achieve an improved tradeoff between fairness and accuracy.…”

Quantifying Feature Contributions to Overall Disparity Using Information Theory

“…However, [2,[12][13][14][15][16][17][18][19] focus on quantifying exempt and non-exempt discrimination (either using observational measures or causal modelling) given a choice of critical features, rather than quantifying the contribution of each individual feature. Alternatively, [33,34] propose information-theoretic techniques to carefully select features for fair decision making. In this work, we focus on explaining the contributions of all the individual features to the overall disparity, even when we may not have access to the exact decision-making mechanism (including scenarios with human-in-the-loop).…”