🤖 AI Summary
To address the lack of actionable intervention mechanisms in multivariate time-series anomaly detection, this paper pioneers the integration of algorithmic recourse into this domain, proposing RecAD—a fully differentiable, end-to-end framework. RecAD enables causal-aware, cross-variable and cross-timestep action recommendations by modeling differentiable time-series perturbations via gradient-based optimization and constraint projection; it incorporates domain knowledge to define the action space and performs cost-weighted recourse search. Experiments on two synthetic benchmarks and one real-world industrial time-series dataset demonstrate that RecAD achieves a 37% average improvement in repair success rate and reduces total action cost by 52%. Its core contributions are: (1) introducing the first algorithmic recourse paradigm specifically designed for time-series anomalies; (2) establishing the first differentiable recourse framework supporting dynamic, multivariate interventions; and (3) empirically validating its effectiveness and practical utility in real-world industrial settings.
📝 Abstract
Anomaly detection in multivariate time series has received extensive study due to the wide spectrum of applications. An anomaly in multivariate time series usually indicates a critical event, such as a system fault or an external attack. Therefore, besides being effective in anomaly detection, recommending anomaly mitigation actions is also important in practice yet under-investigated. In this work, we focus on algorithmic recourse in time series anomaly detection, which is to recommend fixing actions on abnormal time series with a minimum cost so that domain experts can understand how to fix the abnormal behavior. To this end, we propose an algorithmic recourse framework, called RecAD, which can recommend recourse actions to flip the abnormal time steps. Experiments on two synthetic and one real-world datasets show the effectiveness of our framework.
