FMP-AE: A HYBRID APPROACH TO TIME SERIES ANOMALY DETECTION

24 Sept 2024 (modified: 05 Feb 2025)Submitted to ICLR 2025EveryoneRevisionsBibTeXCC BY 4.0
Keywords: Anomaly detection; Matrix Profile; Deep learning;
TL;DR: The FMP-AE model enhances time series anomaly detection by integrating matrix profil with deep learning methods for improving accuracy and efficiency.
Abstract:

Unsupervised anomaly detection in time series presents significant challenges, especially due to the lack of labeled data and the prevalence of highly imbalanced datasets. Traditional statistical and machine learning methods often suffer from low recall and computational inefficiency. While deep learning techniques can automatically extract features, they still struggle with data imbalance. This paper introduces a novel anomaly detection model, Feature map Matrix Profile with an AutoEncoder (FMP-AE), which integrates matrix profile techniques with deep learning. The model uses a 1D-CNN to extract features and compute the matrix profile. A new Matrix Profile loss function is introduced and combined with the Autoencoder's reconstruction loss to enhance anomaly detection. The approach also incorporates a sliding window technique to improve sensitivity to sparse anomalies and increase efficiency. Experimental results on the UCR250 benchmark datasets demonstrate the model's superior performance across multiple metrics, including accuracy, precision, recall, F1-score, and AUC. These results highlight the FMP-AE model's ability to efficiently process large-scale datasets and generalize well across diverse time series domains, offering significant improvements in both detection accuracy and computational efficiency.

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.
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: 3676
Loading

OpenReview is a long-term project to advance science through improved peer review with legal nonprofit status. We gratefully acknowledge the support of the OpenReview Sponsors. © 2025 OpenReview