SafeAR: Safe Algorithmic Recourse by Risk-Aware Policies

TL;DR

SafeAR reframes algorithmic recourse as a finite-horizon, risk-aware decision process by formulating recourse as a finite-horizon MDP and introducing a Greedy Risk-Sensitive Value Iteration (G-RSVI) algorithm. The method incorporates cost uncertainty through mean-variance trade-offs and standard financial risk measures VaR_α and CVaR_α, enabling multiple risk-tolerant recourse policies. Evaluations on the UCI Adult Income and German Credit datasets show that higher risk-aversion (larger β) lowers cost variance and tail risks (VaR/CVaR) at the expense of higher average costs, while highlighting that sparsity and proximity do not guarantee risk safety. The work also analyzes gender-based risk disparities, suggesting that risk exposure in recourse can be unequal across groups, which motivates further fairness considerations in practical deployment.

Abstract

With the growing use of machine learning (ML) models in critical domains such as finance and healthcare, the need to offer recourse for those adversely affected by the decisions of ML models has become more important; individuals ought to be provided with recommendations on actions to take for improving their situation and thus receiving a favorable decision. Prior work on sequential algorithmic recourse -- which recommends a series of changes -- focuses on action feasibility and uses the proximity of feature changes to determine action costs. However, the uncertainties of feature changes and the risk of higher than average costs in recourse have not been considered. It is undesirable if a recourse could (with some probability) result in a worse situation from which recovery requires an extremely high cost. It is essential to incorporate risks when computing and evaluating recourse. We call the recourse computed with such risk considerations as Safe Algorithmic Recourse (SafeAR). The objective is to empower people to choose a recourse based on their risk tolerance. In this work, we discuss and show how existing recourse desiderata can fail to capture the risk of higher costs. We present a method to compute recourse policies that consider variability in cost and connect algorithmic recourse literature with risk-sensitive reinforcement learning. We also adopt measures "Value at Risk" and "Conditional Value at Risk" from the financial literature to summarize risk concisely. We apply our method to two real-world datasets and compare policies with different risk-aversion levels using risk measures and recourse desiderata (sparsity and proximity).

Figures (6)

• Figure 1: Recourse policies for loan approvals. Policy A has only one feature change (low sparsity) but with a high failure rate; Policy B has the lowest expected cost but might result in a situation that costs more to recover; Policy C has a slightly higher expected cost than Policy B but lower variance in cost (risk), which can be considered as a safer Policy.
• Figure 2: Visualizing risks in three recourse policies from Figure \ref{['fig:stable_plan']}. The x-axis indicates the cost, and the line (outcome) thickness indicates the outcome probability.
• Figure 3: Policy visualization Example in German Credit.
• Figure 4: Health insurance premiums recourse (synthetic)
• Figure 5: Moving average of total recourse cost during G-RSEVI updates with maximum episode 10,000 and exponential decay factor 0.9995 for $\epsilon$. Shades indicate $\pm\sigma$ cost range of the moving window.