Dominant Shuffle: An Incredibly Simple but Exceptionally Effective Data Augmentation Method for Time-Series Prediction

ICLR 2025 Conference Submission3630 Authors

24 Sept 2024 (modified: 13 Oct 2024)ICLR 2025 Conference SubmissionEveryoneRevisionsBibTeXCC BY 4.0
Keywords: time series prediction, data augmentation, deep learning
TL;DR: A simple yet efficient frequency domain data augmentation method for time series prediction.
Abstract: Frequency-domain data augmentation (DA) has shown strong performance in time-series prediction due to its ability to preserve data-label consistency. However, we observed that existing frequency-domain augmentations introduce excessive variability, leading to out-of-distribution samples that may be harmful to model performance. To address this, we introduced two simple modifications to frequency-domain DA. First, we limit perturbations to dominant frequencies with larger magnitudes, which capture the main periodicities and trends of the signal. Second, instead of using complicated random perturbations, we simply shuffle the dominant frequency components, which preserves the original structure while avoiding external noise. With the two simple modifications, we proposed dominant shuffle—a simple yet highly effective data augmentation technique for time-series prediction. Our method is remarkably simple, requiring only a few lines of code, yet exceptionally effective, consistently and significantly improving model performance. Extensive experiments on short-term, long term, few-shot and cold-start prediction tasks with eight state-of-the-art models, nine existing augmentation methods and twelve datasets demonstrate that dominant shuffle consistently boosts model performance with substantial gains, outperforming existing augmentation techniques.
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: 3630
Loading