Attribute Removal (Fairness Through Unawareness)

Description

Attribute removal, also called fairness through unawareness, trains and runs a model without the protected attributes, such as race, gender or age, among its inputs. This meets the unawareness criterion, under which an algorithm counts as fair so long as it does not explicitly use a protected attribute, and it rules out direct discrimination on those attributes. It does not make outcomes fair. Other features can act as proxies, meaning they correlate with a removed attribute and let the model reconstruct it, for example a postcode standing in for ethnicity. For that reason the technique is best seen as the simplest baseline, to be combined with methods that look at outcomes.

Example Use Cases

Fairness

A recruitment agency's data team builds a CV-ranking model for graduate roles and leaves gender, ethnicity and age out of its inputs, so that shortlisting cannot use them directly. It records that education institution and postcode may still carry some of that information and treats this as an open risk, to be handled with other methods.

A bank's credit-risk team removes sex and ethnicity from the inputs to a loan-approval model so that its decisions cannot use them directly, as its lending policy requires. It keeps the two fields in a separate monitoring table, so approval rates can still be compared across groups after launch.

A health-system analytics team builds a postpartum depression risk model from insurance claims and first tries leaving race out of the inputs. It then measures disparate impact between Black and White patients and compares the result with other bias-reduction methods on the same data, to decide whether removal is enough.

Limitations

Proxy features keep the removed attribute's influence in the model. Postcode, school or employment history can predict race, sex or age closely enough for the model to reproduce group differences without seeing the attribute. In a published comparison on Medicaid postpartum depression data, removing race left disparate impact at about 0.6 (1 is parity), less effective than the other methods compared.
Proxies cannot be found by inspecting the feature list, because group membership can be spread across many features, and checking outcomes by group needs the very attributes that were removed. A team that deletes the column outright loses the ability to see whether the problem remains.
Removal alone is weak compliance evidence. Regulators that test for disparate impact or indirect discrimination look at outcomes, not inputs, so a model that never saw the attribute can still fail the test.
Removal can hurt the group it is meant to protect. Where the link between features and outcomes differs by group, dropping the attribute can raise error for the group concerned. Kusner et al. give a case where an 'unaware' predictor perpetuates unfairness, and Dwork et al. note that errors can come from ignoring group membership.

Resources

Research Papers

Fairness Through Awareness

Cynthia Dwork et al.•Apr 20, 2011

We study fairness in classification, where individuals are classified, e.g., admitted to a university, and the goal is to prevent discrimination against individuals based on their membership in some group, while maintaining utility for the classifier (the university). The main conceptual contribution of this paper is a framework for fair classification comprising (1) a (hypothetical) task-specific metric for determining the degree to which individuals are similar with respect to the classification task at hand; (2) an algorithm for maximizing utility subject to the fairness constraint, that similar individuals are treated similarly. We also present an adaptation of our approach to achieve the complementary goal of "fair affirmative action," which guarantees statistical parity (i.e., the demographics of the set of individuals receiving any classification are the same as the demographics of the underlying population), while treating similar individuals as similarly as possible. Finally, we discuss the relationship of fairness to privacy: when fairness implies privacy, and how tools developed in the context of differential privacy may be applied to fairness.

Counterfactual Fairness

Matt J. Kusner et al.•Dec 4, 2017

Machine learning can impact people with legal or ethical consequences when it is used to automate decisions in areas such as insurance, lending, hiring, and predictive policing. In many of these scenarios, previous decisions have been made that are unfairly biased against certain subpopulations, for example those of a particular race, gender, or sexual orientation. Since this past data may be biased, machine learning predictors must account for this to avoid perpetuating or creating discriminatory practices. In this paper, we develop a framework for modeling fairness using tools from causal inference. Our definition of counterfactual fairness captures the intuition that a decision is fair towards an individual if it the same in (a) the actual world and (b) a counterfactual world where the individual belonged to a different demographic group. We demonstrate our framework on a real-world problem of fair prediction of success in law school.

Related Techniques

Name	Description	Assurance Goals
Disparate Impact Remover	Disparate Impact Remover is a pre-processing method that edits the numerical features of a dataset so that they reveal less about a protected attribute such as race or sex. For each feature it compares how the values are distributed within each protected group, then moves every value towards a shared distribution while keeping its rank within its own group, so people near the top of their group stay near the top. The user sets a repair level from 0 (no change) to 1 (each feature is distributed identically in every group). The aim is to avoid disparate impact, where one group's selection rate falls below 80 per cent of the most favoured group's rate. Stronger repair narrows that gap but costs predictive accuracy.	Fairness
Fair Adversarial Networks	Fair Adversarial Networks is a pre-processing method, proposed by Cevora (2020), that rewrites a dataset so that a protected characteristic such as race, sex or age is hard to predict from it, including through proxies such as postcode or occupation. It trains two neural networks against each other. An autoencoder (a network that learns to reproduce its input) rewrites each record while staying as close to the original as it can, and an adversary network tries to predict the protected characteristic from the rewritten records. The autoencoder is penalised both for changing the data and for the adversary's success, so over alternating rounds of training it strips out the information the adversary relies on and changes little else. The output is the rewritten dataset, which any downstream analysis or model can use unchanged, and in the paper success means a freshly trained adversary does no better than always guessing the most common category.	Fairness
Relabelling	Relabelling, also called data massaging, is a pre-processing method proposed by Kamiran and Calders. It changes some of the outcome labels in a training set so that a protected group and everyone else have the same rate of positive outcomes, such as being hired or approved. A ranking model first scores each example by how likely it is to be positive. The method then promotes the highest-scoring negative examples from the disadvantaged group to positive and demotes the lowest-scoring positive examples from the favoured group to negative, in equal numbers. Because the examples closest to the decision boundary change first and the numbers are equal, the overall share of positive labels stays the same and the number of changes is as small as it can be.	Fairness