Actionable Recourse in Linear Classification

By Berk Ustun, Alexander Spangher, Yang Liu. 2018

662 citations recorded by the source corpus.

Classification models are often used to make decisions that affect humans: whether to approve a loan application, extend a job offer, or provide insurance. In such applications, individuals should have the ability to change the decision of the model. When a person is denied a loan by a credit scoring model, for example, they should be able to change the input variables of the model in a way that will guarantee approval. Otherwise, this person will be denied the loan so long as the model is deployed, and -- more importantly --will lack agency over a decision that affects their livelihood. In this paper, we propose to evaluate a linear classification model in terms of recourse, which we define as the ability of a person to change the decision of the model through actionable input variables (e.g., income vs. age or marital status). We present an integer programming toolkit to: (i) measure the feasibility and difficulty of recourse in a target population; and (ii) generate a list of actionable changes for a person to obtain a desired outcome. We discuss how our tools can inform different stakeholders by using them to audit recourse for credit scoring models built with real-world datasets. Our results illustrate how recourse can be significantly affected by common modeling practices, and motivate the need to evaluate recourse in algorithmic decision-making.

Subfields: Artificial Intelligence, Safety Research, AI Ethics, Fairness & Society

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Safety Research

In AI Ethics, Fairness & Society

Citations in the corpus

CERTIFAI: Counterfactual Explanations for Robustness, Transparency, Interpretability, and Fairness of Artificial Intelligence models (2019, methodology): “Methods in [Ustun et al., 2019] and [Russell, 2019] only work for linear models.”
AI Fairness 360: An Extensible Toolkit for Detecting, Understanding, and Mitigating Unwanted Algorithmic Bias (2018, methodology): “Future releases may include methodologies such as fine-grained localization of bias (we describe the approach herein), actionable recourse analysis (Ustun et al., 2018), and counterfactual fairness (Wachter et al.”
On the Fairness of Causal Algorithmic Recourse (2020, background): “Ustun, Spangher, and Liu (2019) refer to a person’s ability to change the decision of a model by altering actionable 3We use v when there is an explicit distinction between the protected attribute and other features (in the context of fairness) and x otherwise (in the context of…”
Algorithmic recourse under imperfect causal knowledge: a probabilistic approach (2020, background): “To address this limitation, the recently proposed framework of algorithmic recourse focuses instead on the actions an individual can perform to achieve a more favourable outcome [44].”
Getting a CLUE: A Method for Explaining Uncertainty Estimates (2020, background): “Telling a person that they would have been approved for a loan had their age been − 10 is of very little use. et To right this, recent works deﬁne linear constraints on explanations (Ustun al., 2019; Sharma et al., 2020).”
Toward Trustworthy AI Development: Mechanisms for Supporting Verifiable Claims (2020, neutral): “, bank loans) [77];”
ViCE: visual counterfactual explanations for machine learning models (2020, background): “It is also worth noticing that the visualization interface can accommodate any other counterfactual generation methods [9, 16, 21, 22].”
Explainable machine learning in deployment (2019, background): “exist [63, 65], they either work for specific model-types or are not black-box in nature.”
Manipulating and Measuring Model Interpretability (2018, background): “[103] Berk Ustun, Alexander Spangher, and Yang Liu.”
Efficient Search for Diverse Coherent Explanations (2019, background): “Finally, concurrent with this work, Ustun et al. (2019), have also proposed the generation of diverse counterfactuals using mixed in-teger programmes for linear models.”
A survey of algorithmic recourse: definitions, formulations, solutions, and prospects (2020, methodology): “Finally, the need to analyze recourse on a sub-population level may arise due to uncertainty in assumptions [93] or as an intentional study of other properties of the system, e.g., fairness [42, 77, 91, 153, 171], which we explore further below.”
Should ChatGPT be Biased? Challenges and Risks of Bias in Large Language Models (2023, background): “Biased models may unfairly penalize certain groups, such as minority communities or individuals with lower socioeconomic status, by assigning them lower credit scores or denying them access to loans and financial services based on biased assumptions [104, 105, 106].”
Counterfactual Explanations and Algorithmic Recourses for Machine Learning: A Review (2020, neutral): “(a) Sparsity: Among the collected articles, methods using solvers explicitly constrain sparsity [182, 331], black-box methods constrain L0 norm of counterfactual and the input dat-apoint [75, 210].”
The philosophical basis of algorithmic recourse (2020, background): “In [31], this problem is addressed by allowing the system to return multiple “flipsets” – ways to take action to remedy an unfavorable outcome – as a recognition of this potential lack of knowledge of the true cost”
The hidden assumptions behind counterfactual explanations and principal reasons (2019, background): “And indeed, some scholars have suggested as much [44].”