A-CBFI：面向稀疏因果算法追索的处方式干预诊断与交互感知升级方法
Actionable CBFI: Prescriptive Interventional Diagnosis and Interaction-Aware Escalation for Sparse Causal Algorithmic Recourse
AI 导读
Sejong Oh 提出 Actionable Case-Based Feature Importance（A-CBFI），将 Localized-CBFI 的描述性分解转化为结构因果模型上的处方式干预诊断，存在单动作时优先开最便宜的单动作，不够时才搜索联合动作。
正文 · 原文
Abstract:Algorithmic recourse tells an individual which feasible actions would change an unfavourable prediction. Feature attribution explains a prediction, but the most important causes are often not the best levers. We propose Actionable Case-Based Feature Importance (A-CBFI), which turns the descriptive decomposition of Localized-CBFI into a prescriptive, interventional diagnosis on a structural causal model. A-CBFI prescribes the cheapest single action when one exists and searches joint actions only when no single action is enough. We show that interventional interaction also arises from path blocking and shared descendants, which model-only interaction indices cannot see, and that averaged interaction scores cancel where additive targeting fails. On ten tabular benchmarks (6,657 recourse queries), A-CBFI was never significantly less successful than an untargeted causal beam search, prescribed fewer direct actions (1.10 vs 1.38) with fewer model evaluations (119 vs 150), and offered substitutable single actions for 68.2% of the queries a single action solves. Success gains appeared mainly where causes are immutable and single actions are insufficient.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| MSC classes: | 62R07 |
| ACM classes: | I.2.6 |
| Cite as: | arXiv:2608.27821 [cs.LG] |
| (or arXiv:2608.27821v3 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2608.27821 arXiv-issued DOI via DataCite |
Submission history
From: Sejong Oh [view email]
[v1]
Fri, 28 Aug 2026 01:40:43 UTC (2,116 KB)
[v2]
Tue, 8 Sep 2026 00:19:11 UTC (2,133 KB)
[v3]
Tue, 27 Oct 2026 06:42:17 UTC (1,292 KB)
来源：arXiv cs.AI · arxiv.org