Graph-Attention Diffusion for Enhanced Multivariate Time-Series Anomaly Detection
Multivariate time-series anomaly detection is a complex task that requires capturing temporal and spatial correlations. Recently, among the unsupervised methods, diffusion models have attracted increased attention among researchers for addressing this particular task. However, spatial information of...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2024-01-01
|
| Series: | IEEE Open Journal of the Industrial Electronics Society |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10755103/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832590324849967104 |
|---|---|
| author | Vadim Lanko Ilya Makarov |
| author_facet | Vadim Lanko Ilya Makarov |
| author_sort | Vadim Lanko |
| collection | DOAJ |
| description | Multivariate time-series anomaly detection is a complex task that requires capturing temporal and spatial correlations. Recently, among the unsupervised methods, diffusion models have attracted increased attention among researchers for addressing this particular task. However, spatial information often remains underutilized or overlooked in existing models. In this article, we propose a novel reconstruction-based approach that enhances normal pattern learning through data masking and leverages diffusion models to capture both temporal and spatial interrelations via graph-attention layers. To address the problem of overgeneralization, where anomalous points are reconstructed too well, potentially abnormal points are initially masked based on the reconstruction error produced by the autoencoder. The masked time-series data is then corrupted by noise and reconstructed back by the diffusion model that removes noise in a step-by-step manner. Evaluation on four datasets from various sources demonstrates the effectiveness of our approach, achieving an average <inline-formula><tex-math notation="LaTeX">$F1$</tex-math></inline-formula>-score of 96.51%, outperforming many existing baselines. The ablation study estimates the contribution of each of the key components of the model to the score. |
| format | Article |
| id | doaj-art-4e13e7883ba640cdac648fdefb876bf1 |
| institution | Kabale University |
| issn | 2644-1284 |
| language | English |
| publishDate | 2024-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Open Journal of the Industrial Electronics Society |
| spelling | doaj-art-4e13e7883ba640cdac648fdefb876bf12025-01-24T00:02:15ZengIEEEIEEE Open Journal of the Industrial Electronics Society2644-12842024-01-0151353136410.1109/OJIES.2024.350101410755103Graph-Attention Diffusion for Enhanced Multivariate Time-Series Anomaly DetectionVadim Lanko0https://orcid.org/0009-0000-1426-120XIlya Makarov1https://orcid.org/0000-0002-3308-8825Moscow Institute of Physics and Technology, Moscow, RussiaArtificial Intelligence Research Institute (AIRI), Moscow, RussiaMultivariate time-series anomaly detection is a complex task that requires capturing temporal and spatial correlations. Recently, among the unsupervised methods, diffusion models have attracted increased attention among researchers for addressing this particular task. However, spatial information often remains underutilized or overlooked in existing models. In this article, we propose a novel reconstruction-based approach that enhances normal pattern learning through data masking and leverages diffusion models to capture both temporal and spatial interrelations via graph-attention layers. To address the problem of overgeneralization, where anomalous points are reconstructed too well, potentially abnormal points are initially masked based on the reconstruction error produced by the autoencoder. The masked time-series data is then corrupted by noise and reconstructed back by the diffusion model that removes noise in a step-by-step manner. Evaluation on four datasets from various sources demonstrates the effectiveness of our approach, achieving an average <inline-formula><tex-math notation="LaTeX">$F1$</tex-math></inline-formula>-score of 96.51%, outperforming many existing baselines. The ablation study estimates the contribution of each of the key components of the model to the score.https://ieeexplore.ieee.org/document/10755103/Diffusiongraph neural network (GNN)time-series anomaly detection |
| spellingShingle | Vadim Lanko Ilya Makarov Graph-Attention Diffusion for Enhanced Multivariate Time-Series Anomaly Detection IEEE Open Journal of the Industrial Electronics Society Diffusion graph neural network (GNN) time-series anomaly detection |
| title | Graph-Attention Diffusion for Enhanced Multivariate Time-Series Anomaly Detection |
| title_full | Graph-Attention Diffusion for Enhanced Multivariate Time-Series Anomaly Detection |
| title_fullStr | Graph-Attention Diffusion for Enhanced Multivariate Time-Series Anomaly Detection |
| title_full_unstemmed | Graph-Attention Diffusion for Enhanced Multivariate Time-Series Anomaly Detection |
| title_short | Graph-Attention Diffusion for Enhanced Multivariate Time-Series Anomaly Detection |
| title_sort | graph attention diffusion for enhanced multivariate time series anomaly detection |
| topic | Diffusion graph neural network (GNN) time-series anomaly detection |
| url | https://ieeexplore.ieee.org/document/10755103/ |
| work_keys_str_mv | AT vadimlanko graphattentiondiffusionforenhancedmultivariatetimeseriesanomalydetection AT ilyamakarov graphattentiondiffusionforenhancedmultivariatetimeseriesanomalydetection |