Algorithmic Recourse for Long-Term Improvement

Algorithmic recourse aims to provide a recourse action for altering an unfavorable prediction given by a model into a favorable one (e.g., loan approval). In practice, it is also desirable to ensure that an action makes the real-world outcome better (e.g., loan repayment). We call this requirement improvement. Unfortunately, existing methods cannot ensure improvement unless we know the true oracle. To address this issue, we propose a framework for suggesting improvement-oriented actions from a long-term perspective. Specifically, we introduce a new online learning task of assigning actions to a given sequence of instances. We assume that we can observe delayed feedback on whether the past suggested action achieved improvement. Using the feedback, we estimate an action that can achieve improvement for each instance. To solve this task, we propose two approaches based on contextual linear bandit and contextual Bayesian optimization. Experimental results demonstrated that our approaches could assign improvement-oriented actions to more instances than the existing methods.

Machine learning models are increasingly used to make important decisions, like loan approvals. When a person is denied a loan, a bank is required to inform the person of what "actions" they can take to get approved in the future. This is where "algorithmic recourse" comes in, suggesting changes a person can make to achieve a desired outcome. However, current methods mainly focus on changing the model's prediction, without considering whether these changes actually improve the person's real-world situation (e.g., their ability to repay the loan). This research introduces a new approach that focuses on long-term improvement. The core idea is to learn from past experiences: observing whether suggested actions truly led to better results, and this feedback refines future suggestions. To achieve this, we developed new algorithms to learn which actions are most likely to lead to improvement. We utilize a well-established learning framework called bandit algorithms. By focusing on real-world outcomes and learning from them, our system adapts to provide more effective and reliable guidance. This leads to more helpful decision-making, empowering individuals with actions that genuinely benefit them in the long run.