When Will It Fail?: Anomaly to Prompt for Forecasting Future Anomalies in Time Series

Min-Yeong Park, Won-Jeong Lee, Seong Tae Kim, Gyeong-Moon Park

Keywords: Time Series, Time Series Forecasting, Anomaly Detection

Abstract: Recently, time series forecasting, which predicts future signals, and time series anomaly detection, which identifies abnormal signals in given data, have achieved impressive success. However, in the real world, merely forecasting future signals or detecting anomalies in existing signals is not sufficiently informative to prevent potential system breakdowns, which lead to huge costs and require intensive human labor. In this work, we tackle a challenging and under-explored problem of time series anomaly prediction. In this scenario, the models are required to forecast the upcoming signals while considering anomaly points to detect them. To resolve this challenging task, we propose a simple yet effective framework, Anomaly to Prompt (A2P), which is jointly trained via the forecasting and anomaly detection objectives while sharing the feature extractor for better representation. On top of that, A2P leverages Anomaly-Aware Forecasting (AAF), which derives the anomaly probability by random anomaly injection to forecast abnormal time points. Furthermore, we propose Synthetic Anomaly Prompting (SAP) for more robust anomaly detection by enhancing the diversity of abnormal input signals for training anomaly detection model. As a result, our model achieves state-of-the-art performances on seven real-world datasets, proving the effectiveness of our proposed framework A2P for a new time series anomaly prediction task.

Primary Area: learning on time series and dynamical systems

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 6191