Evaluating Language Model Features for Zero-Shot Time-Series Anomaly Detection Under Isolation Assumption
Organizations: University of Luxembourg
Abstract
Time-Series Anomaly Detection (TSAD) is a fundamentally challenging task: it is unsupervised, requires high local precision capabilities, and suffers from limited labelled data. Recent work has explored adapting Large Language Models (LLMs) to time-series via cross-modal transfer. However, existing approaches are poorly suited to Multivariate TSAD (MTSAD), as tokenisation often obscures fine-grained temporal and variables structures, and most methods optimise auxiliary objectives such as forecasting or reconstruction. We posit that these surrogate objectives become unreliable as model capacity increases. Highly expressive models risk reconstructing or forecasting anomalies, undermining the foundations of both strategies. Instead, we propose to use LLMs purely as feature extractors and directly analyse their induced feature spaces. In this work, investigate the ability of LLMs to provide useful representations for MTSADs in a zero-shot setting through classification. To this end, we study the separability of the extracted features, under the assumption that anomalous features should be isolated from normal ones. We analyse their temporal and contextual discriminative ability using simple step-wise tokenisation schemes, with one time-step per token, which preserve fine-grained local information. Their usability for MTSAD is assessed using an unsupervised Isolation Forest, enabling a direct evaluation of the resulting representation quality. We further investigate whether the observed performance can be attributed to the language-related pre-training of the LLMs or arises independently of it. Our results provide empirical insight into the ability of LLMs to represent multivariate time-series for anomaly detection and highlight broader applications of cross-modal foundation models.
Figures & tables
| Model | PSM | PUMP | SWaT | WaDi | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | |||
| Trained | MSE | AnomalyTransformer | 68.12 00.40 | 78.97 00.64 | 64.00 00.85 | 22.00 00.07 | 79.56 00.06 | 49.09 00.75 | 69.97 01.27 | 80.88 00.10 | 71.69 02.15 | 29.27 00.01 | 53.03 00.10 | 14.99 00.14 |
| GPT4TS | 52.76 00.11 | 62.05 00.07 | 48.65 00.23 | 07.35 00.00 | 21.02 00.03 | 18.66 00.05 | 10.92 00.02 | 22.12 00.02 | 10.33 00.13 | 41.37 00.00 | 55.82 00.01 | 14.07 00.01 | ||
| TimesNet | 51.83 00.10 | 61.55 00.06 | 47.72 00.26 | 07.38 00.01 | 21.15 00.06 | 18.83 00.17 | 10.72 00.02 | 21.83 00.07 | 09.39 01.46 | 41.38 00.01 | 55.95 00.07 | 14.05 00.01 | ||
| Pre-Trained | Chronos ( ) | 44.17 | 57.55 | 44.58 | 09.04 | 27.75 | 12.90 | 09.63 | 26.94 | 02.04 | 06.68 | 48.62 | 13.14 | |
| Chronos ( ) | 43.84 | 59.86 | 53.37 | 09.51 | 27.84 | 16.88 | 09.44 | 26.96 | 05.49 | 06.92 | 50.40 | 12.88 | ||
| Model | PSM | PUMP | SWaT | WaDi | |||||||||
| AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | ||
| GPT | 32.92 0.41 | 54.08 0.94 | 35.81 4.44 | 17.14 1.08 | 67.92 1.27 | 32.98 2.14 | 53.10 6.01 | 67.71 0.76 | 55.55 7.40 | 15.22 6.46 | 43.17 2.77 | 12.53 0.16 | |
| 33.70 | 54.86 | 37.13 | 16.44 | 66.86 | 31.78 | 44.74 | 65.79 | 49.47 | 12.83 | 41.44 | 12.14 | ||
| *36.13 | *56.84 | *39.24 | *18.72 | *73.40 | *38.44 | *46.66 | *67.07 | *48.91 | *16.74 | *43.23 | *12.37 | ||
| - | - | 26.93 | 57.24 | 32.63 | *29.56 | *60.13 | *34.74 | ||||||
| Mistral | 32.88 0.89 | 56.78 0.92 | 37.38 5.91 | 18.47 0.78 | 74.92 1.36 | 38.51 0.99 | 26.36 10.52 | 64.59 3.06 | 36.03 4.84 | 21.40 3.10 | 45.40 1.33 | 11.96 0.32 | |
Appendix figures & tables31 assets
Supplementary material from the paper’s appendix.
Appendix
| Architecture | HuggingFace ( Wolf et al., 2019 ) Version | Number of Parameters | Embedding Size | Multimodal Training | Vocabulary Size |
| AnomalyTransformer ( Xu et al., 2022 ) | - | 5,002,387 | 512 | ✗ | - |
| GPT4TS ( Zhou et al., 2023 ) | - | 81,960,828 | 768 | ✗ | - |
| TimesNet ( Wu et al., 2023 ) | - | 28,150,972 | 128 | ✗ | - |
| Chronos ( Ansari et al., 2025 ) | amazon/chronos-2 | 119,477,664 | ✗ | - | |
| MOMENT (Small) ( Goswami et al., 2024 ) | AutonLab/MOMENT-1-small | 35,341,512 | 512 | ✗ | - |
| MOMENT (Base) ( Goswami et al., 2024 ) | AutonLab/MOMENT-1-base | 109,641,608 | 768 | ✗ | - |
| Dataset | Features | Train Set | Test Set | Anomaly % | Anomalous Sequences | |||
|---|---|---|---|---|---|---|---|---|
| Length | Num. | |||||||
| Min. | Max. | Avg. | ||||||
| SWaT ( Abdulaal et al., 2021 ) | 51 | 495,000 | 449,919 | 12.14 | 101 | 35,900 | 1,560.60 | 35 |
| WaDi ( Ahmed et al., 2017 ) | 123 | 784,537 | 172,801 | 5.77 | 88 | 1,741 | 712.64 | 14 |
| PUMP ( Feng and Tian, 2021 ) | 44 | 76,901 | 143,401 | 10.05 | 42 | 8,391 | 2,401.33 | 6 |
| PSM ( Abdulaal et al., 2021 ) | 25 | 132,481 | 87,841 | 27.76 | 1 | 8,861 | 338.625 | 72 |
| Dataset | Stride | Test Size (# Windows) | Anomaly % w.r.t. Window Size | ||||||
| 1 | 4 | 8 | 16 | 32 | 64 | 128 | |||
| PSM | 1 | 87841 | 27.76 | 28.00 | 28.31 | 28.87 | 29.91 | 31.71 | 35.06 |
| 4 | 21961 | 27.99 | |||||||
| 8 | 10981 | 28.37 | |||||||
| 16 | 5491 | 28.88 | |||||||
| 32 | 2746 | 30.01 | |||||||
| Dataset | Stride | Test Size (# Windows) | Anomaly Window Coverage (%) w.r.t. Window Size | ||||||
| 1 | 4 | 8 | 16 | 32 | 64 | 128 | |||
| PSM | 1 | 87841 | 100.0 | 99.12 | 98.05 | 96.09 | 92.77 | 87.45 | 79.00 |
| 4 | 21961 | 99.17 | |||||||
| 8 | 10981 | 97.85 | |||||||
| 16 | 5491 | 96.09 | |||||||
| 32 | 2746 | 92.48 | |||||||
| Dataset | Stride | Test Size (# Windows) | Anomaly Segment Centroid (% from left to right) | ||||||
| 1 | 4 | 8 | 16 | 32 | 64 | 128 | |||
| PSM | 1 | 87841 | 00.00 | 50.00 | 50.00 | 50.01 | 50.01 | 50.05 | 50.06 |
| 4 | 21961 | 49.94 | |||||||
| 8 | 10981 | 49.98 | |||||||
| 16 | 5491 | 50.11 | |||||||
| 32 | 2746 | 50.16 | |||||||
| Model | PSM | PUMP | SWaT | WaDi | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | |
| AnomalyTransformer | 68.25 | 78.76 | 63.98 | 21.98 | 79.65 | 48.90 | 69.13 | 80.83 | 72.44 | 29.40 | 52.43 | 14.83 |
| GPT4TS | 52.87 | 61.68 | 49.18 | 07.64 | 20.79 | 21.87 | 10.77 | 21.95 | 10.19 | 40.91 | 55.17 | 14.17 |
| TimesNet | 52.25 | 61.44 | 48.74 | 07.63 | 20.85 | 21.89 | 10.60 | 21.77 | 10.04 | 40.87 | 55.14 | 14.19 |
| MOMENT (Small) | 52.53 | 64.94 | 45.35 | 12.43 | 43.94 | 20.43 | 12.19 | 27.95 | 10.57 | 19.05 | 51.97 | 14.02 |
| MOMENT (Base) | 66.81 | 70.80 | 54.92 | 15.33 | 44.56 | 26.28 | 12.18 | 28.03 | 05.09 | 18.48 | 52.11 | 14.08 |
| Adapter | Window Size | PSM | PUMP | SWaT | WaDi | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | ||
| Raw | 4 | 18.63 0.47 | 26.03 1.58 | 0.00 0.00 | 5.90 0.18 | 16.76 2.64 | 14.35 8.02 | 6.82 0.09 | 13.40 0.32 | 0.43 0.96 | 3.81 0.18 | 24.45 2.42 | 6.38 4.85 |
| 8 | 19.01 0.24 | 26.50 1.17 | 0.00 0.00 | 5.84 0.13 | 15.86 2.13 | 14.28 7.98 | 6.88 0.05 | 13.93 0.39 | 0.50 1.12 | 3.70 0.03 | 25.99 0.52 | 9.01 5.04 | |
| 16 | 19.17 0.37 | 25.49 1.19 | 0.00 0.00 | 17.76 3.06 | 60.83 2.87 | 27.38 3.69 | 40.44 4.11 | 63.77 1.52 | 39.87 3.67 | 3.69 0.06 | 25.77 1.21 | 8.97 5.01 | |
| 32 | 31.64 1.32 | 45.53 2.25 | 10.73 2.10 | 14.84 0.72 | 58.19 0.92 | 24.32 0.70 | 42.12 4.74 | 62.96 1.11 | 42.92 3.68 | 4.08 0.14 | 27.87 0.80 | 5.93 5.76 | |
| 64 | 31.97 1.20 | 43.17 1.65 | 11.48 4.15 | 20.34 2.19 | 58.85 0.95 | 26.49 0.70 | 25.23 7.95 | 56.89 2.58 | 30.22 4.81 | 4.23 0.10 | 28.74 0.81 | 9.50 5.31 | |
| Model | Window Size | PSM | PUMP | SWaT | WaDi | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | ||
| GPT | 4 | 33.32 0.91 | 60.38 0.66 | 45.18 1.77 | 13.59 1.02 | 64.02 1.76 | 22.49 4.88 | 10.04 1.62 | 32.15 2.98 | 20.83 2.94 | 12.25 8.15 | 41.68 1.68 | 9.03 5.05 |
| 8 | 34.65 0.67 | 58.35 1.24 | 44.22 2.29 | 18.60 10.30 | 65.87 1.59 | 24.75 2.63 | 11.73 2.04 | 42.58 5.04 | 18.86 1.97 | 14.60 3.59 | 42.36 1.68 | 11.63 0.35 | |
| 16 | 33.37 0.65 | 56.59 1.30 | 45.04 1.13 | 16.70 0.80 | 70.37 0.68 | 27.79 1.13 | 21.90 13.48 | 53.24 8.58 | 11.98 8.21 | 18.81 5.59 | 44.69 1.65 | 12.02 0.50 | |
| 32 | 33.17 1.60 | 54.03 1.23 | 41.75 3.72 | 29.65 10.34 | 72.64 1.36 | 36.04 1.70 | 14.94 0.86 | 46.94 2.28 | 18.49 1.61 | 23.54 3.15 | 45.88 0.92 | 12.15 0.39 | |
| 64 | 35.49 1.03 | 54.51 0.63 | 40.73 1.63 | 22.00 1.80 | 72.69 0.95 | 37.77 4.44 | 15.45 0.50 | 47.84 2.35 | 26.44 3.51 | 24.87 0.05 | 44.75 0.60 | 12.20 0.20 | |
| Model | Window Size | PSM | PUMP | SWaT | WaDi | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | ||
| GPT | 4 | 20.84 0.99 | 33.81 2.50 | 0.00 0.00 | 7.63 0.84 | 39.36 6.42 | 17.94 0.09 | 6.75 0.15 | 10.44 0.95 | 8.79 11.83 | 3.66 0.08 | 24.65 1.11 | 4.43 6.07 |
| 8 | 22.17 1.01 | 36.18 2.68 | 0.40 0.89 | 13.36 2.04 | 57.36 2.34 | 21.96 1.53 | 10.96 1.57 | 44.60 4.46 | 19.46 4.43 | 3.76 0.01 | 27.22 0.47 | 11.25 0.06 | |
| 16 | 27.33 1.50 | 42.40 1.81 | 5.77 1.73 | 17.82 2.98 | 61.13 2.05 | 27.54 3.17 | 41.87 2.22 | 63.81 0.96 | 40.37 2.29 | 3.69 0.06 | 25.69 1.31 | 8.97 5.02 | |
| 32 | 29.86 0.96 | 44.66 1.49 | 6.91 2.21 | 14.87 1.35 | 57.44 1.55 | 24.31 1.02 | 42.42 8.58 | 62.46 2.16 | 42.18 7.82 | 4.12 0.12 | 28.29 0.74 | 5.25 5.83 | |
| 64 | 31.83 1.24 | 43.02 2.22 | 13.04 5.52 | 20.87 1.52 | 59.42 1.45 | 28.30 1.91 | 28.34 11.60 | 57.24 4.18 | 33.88 8.84 | 4.20 0.08 | 28.48 0.51 | 7.12 6.50 | |
| Model | Window Size | PSM | PUMP | SWaT | WaDi | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | ||
| GPT | 4 | 33.93 0.27 | 56.15 1.66 | 38.26 3.59 | 12.14 0.45 | 60.44 1.31 | 24.42 1.35 | 37.16 8.51 | 64.33 1.15 | 36.54 2.26 | 11.64 3.24 | 37.53 0.83 | 11.26 0.09 |
| 8 | 33.72 0.91 | 52.53 0.98 | 27.90 4.73 | 13.32 0.23 | 63.37 0.68 | 26.51 1.44 | 22.50 12.47 | 61.20 3.30 | 34.02 6.41 | 10.51 5.69 | 37.37 1.91 | 11.38 0.33 | |
| 16 | 35.05 1.86 | 53.20 1.91 | 27.40 12.40 | 14.39 0.41 | 64.49 0.94 | 27.36 0.81 | 38.09 14.58 | 64.83 2.38 | 43.27 8.80 | 11.95 4.66 | 38.96 1.22 | 11.66 0.32 | |
| 32 | 34.85 1.40 | 54.25 0.43 | 39.97 4.84 | 16.11 1.21 | 66.35 0.78 | 30.62 1.24 | 30.21 13.07 | 62.89 2.52 | 39.30 4.32 | 10.46 4.60 | 38.81 0.83 | 11.95 0.17 | |
| 64 | 37.37 0.77 | 53.54 1.07 | 35.42 10.79 | 14.74 0.50 | 64.35 1.04 | 29.29 2.08 | 18.05 4.11 | 59.62 2.84 | 32.14 4.08 | 15.63 5.12 | 40.63 2.71 | 12.14 0.22 | |
| Model | Window Size | PSM | PUMP | SWaT | WaDi | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | ||
| GPT | 4 | 32.88 0.60 | 56.90 0.84 | 40.03 1.31 | 18.42 11.41 | 64.37 2.81 | 24.55 0.73 | 11.04 1.62 | 40.64 5.46 | 22.58 2.44 | 12.84 0.76 | 42.10 1.60 | 2.20 4.92 |
| 8 | 34.22 0.45 | 57.96 0.86 | 43.42 2.20 | 18.76 10.42 | 66.97 1.88 | 27.15 0.43 | 14.32 1.05 | 43.96 2.36 | 17.44 1.37 | 15.03 3.70 | 42.64 1.88 | 11.07 0.81 | |
| 16 | 31.61 0.62 | 53.52 1.35 | 38.46 8.20 | 17.64 0.95 | 71.54 0.82 | 31.08 3.49 | 14.64 1.42 | 44.46 2.10 | 17.67 1.17 | 17.68 6.18 | 44.60 1.67 | 11.92 0.35 | |
| 32 | 33.03 0.80 | 53.86 1.63 | 40.42 5.97 | 19.21 0.68 | 71.59 0.64 | 31.81 1.98 | 15.67 0.58 | 49.11 1.63 | 19.30 0.26 | 23.93 1.28 | 44.95 0.34 | 11.73 0.09 | |
| 64 | 36.10 0.60 | 55.15 1.42 | 44.08 5.15 | 24.11 7.86 | 71.65 0.98 | 34.67 2.80 | 16.13 0.50 | 50.53 1.94 | 24.62 5.83 | 23.43 3.04 | 43.42 0.78 | 12.05 0.15 | |
| Model | Window Size | PSM | PUMP | SWaT | WaDi | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | ||
| GPT | 4 | 32.41 0.69 | 57.64 1.64 | 39.68 4.47 | 15.15 0.30 | 65.96 0.65 | 30.64 0.98 | 36.39 8.54 | 65.69 1.11 | 45.21 3.51 | 11.46 4.23 | 39.86 1.69 | 11.18 0.19 |
| 8 | 32.82 1.18 | 57.45 1.61 | 41.72 5.00 | 16.12 0.84 | 67.45 0.62 | 31.73 1.55 | 46.80 6.62 | 66.62 0.86 | 50.09 3.62 | 10.51 6.86 | 38.98 2.43 | 11.54 0.29 | |
| 16 | 32.54 0.62 | 55.82 1.14 | 37.43 2.91 | 16.24 0.35 | 67.49 0.40 | 31.20 1.32 | 44.14 14.42 | 66.20 1.97 | 49.00 8.70 | 9.87 4.44 | 40.30 1.26 | 11.56 0.18 | |
| 32 | 32.92 0.41 | 54.08 0.94 | 35.81 4.44 | 17.14 1.08 | 67.92 1.27 | 32.98 2.14 | 53.10 6.01 | 67.71 0.76 | 55.55 7.40 | 15.22 6.46 | 43.17 2.77 | 12.53 0.16 | |
| 64 | 34.90 0.47 | 53.27 0.96 | 34.07 0.82 | 17.29 2.22 | 66.76 0.95 | 32.78 1.91 | 49.43 8.92 | 66.21 1.26 | 51.76 9.00 | 13.87 6.10 | 42.50 3.11 | 12.67 0.35 | |
| Model | Window Size | PSM | PUMP | SWaT | WaDi | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | ||
| Linear | 4 | 22.18 1.16 | 36.07 2.22 | 1.95 2.45 | 9.14 0.97 | 46.68 2.60 | 10.23 9.40 | 7.16 0.12 | 17.95 1.30 | 13.06 11.92 | 3.77 0.04 | 27.31 1.01 | 7.04 5.60 |
| 8 | 22.53 0.60 | 37.73 1.49 | 2.26 2.34 | 14.24 2.00 | 58.57 2.68 | 25.34 4.16 | 22.26 4.34 | 56.25 2.44 | 27.68 3.31 | 3.66 0.05 | 24.96 1.14 | 4.50 6.16 | |
| 16 | 27.12 0.65 | 42.40 0.85 | 5.73 2.05 | 13.03 0.49 | 55.13 0.57 | 22.59 1.06 | 47.51 3.31 | 65.68 0.58 | 47.66 2.30 | 3.70 0.07 | 26.01 1.30 | 11.22 0.03 | |
| 32 | 31.51 1.52 | 45.80 3.95 | 12.36 5.66 | 11.95 0.62 | 51.44 1.69 | 20.53 0.74 | 58.92 0.81 | 68.46 0.47 | 59.20 1.97 | 4.28 0.30 | 28.62 0.62 | 6.47 3.61 | |
| 64 | 35.38 1.32 | 45.99 1.88 | 15.15 4.23 | 10.49 0.69 | 47.83 1.70 | 17.13 2.06 | 59.35 1.82 | 68.20 0.63 | 63.28 3.83 | 4.78 0.33 | 32.69 0.98 | 4.76 2.12 | |
| Model | Window Size | PSM | PUMP | SWaT | WaDi | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | ||
| GPT | 4 | 32.95 1.22 | 57.44 1.80 | 38.27 5.03 | 18.42 0.46 | 74.88 0.76 | 38.92 1.29 | 34.47 8.91 | 65.79 1.80 | 38.29 3.90 | 16.55 0.40 | 41.37 1.01 | 11.55 0.43 |
| 8 | 34.33 0.95 | 58.02 1.08 | 41.70 5.20 | 19.47 0.34 | 76.99 0.45 | 40.80 1.50 | 17.63 3.93 | 60.50 2.55 | 30.11 2.50 | 17.64 2.32 | 42.84 1.21 | 11.91 0.17 | |
| 16 | 35.23 1.82 | 55.55 0.61 | 34.54 6.59 | 19.75 0.18 | 77.05 0.32 | 40.58 1.87 | 18.69 3.51 | 61.58 2.68 | 31.92 4.57 | 15.98 4.28 | 43.35 1.05 | 12.25 0.37 | |
| 32 | 35.36 1.15 | 54.04 0.45 | 30.18 1.52 | 20.09 0.35 | 76.72 0.58 | 42.22 1.11 | 30.66 14.25 | 63.94 3.27 | 36.35 7.92 | 18.28 2.91 | 44.77 0.88 | 12.66 0.31 | |
| 64 | 37.58 1.40 | 52.75 0.84 | 30.10 6.15 | 19.13 0.45 | 74.86 0.76 | 40.74 1.11 | 26.53 12.91 | 62.40 3.41 | 33.70 7.38 | 22.22 2.90 | 46.10 1.54 | 12.99 0.28 | |
| Model | Window Size | PSM | PUMP | SWaT | WaDi | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | ||
| GPT | 4 | 30.53 2.59 | 53.35 2.89 | 39.19 4.85 | 9.52 1.75 | 49.58 9.81 | 22.27 2.50 | 12.73 3.29 | 42.56 11.03 | 18.05 5.01 | 9.10 4.18 | 42.32 3.89 | 11.39 0.35 |
| 8 | 31.46 2.89 | 51.02 5.31 | 32.91 10.05 | 12.58 2.26 | 61.43 6.35 | 26.20 5.40 | 13.73 1.80 | 48.18 6.49 | 6.58 2.07 | 11.89 6.45 | 44.39 1.24 | 11.24 0.98 | |
| 16 | 30.44 0.87 | 51.57 1.56 | 35.87 4.82 | 12.22 3.17 | 58.33 7.25 | 25.86 6.19 | 13.55 2.63 | 49.45 6.07 | 12.29 7.65 | 22.57 3.24 | 44.18 1.60 | 11.55 0.50 | |
| 32 | 30.22 2.80 | 51.03 3.87 | 43.02 4.61 | 10.54 1.77 | 53.27 7.56 | 22.22 3.28 | 25.43 24.03 | 56.79 11.12 | 3.53 2.38 | 22.09 3.77 | 44.87 2.16 | 12.11 0.55 | |
| 64 | 34.94 4.14 | 51.78 2.92 | 43.06 6.66 | 11.72 2.37 | 57.75 6.17 | 24.25 4.18 | 26.01 21.81 | 58.94 8.98 | 14.76 17.49 | 24.80 0.09 | 43.76 1.10 | 12.11 0.15 | |
| Model | Window Size | PSM | PUMP | SWaT | WaDi | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | AUPR | AUROC | F 1 | ||
| GPT | 4 | 34.64 0.63 | 60.26 1.54 | 44.23 3.25 | 17.78 1.40 | 73.21 3.40 | 36.86 2.89 | 48.45 5.62 | 68.18 0.95 | 48.31 4.69 | 13.40 2.90 | 40.60 1.45 | 11.70 0.42 |
| 8 | 34.88 1.52 | 58.75 1.34 | 42.95 2.76 | 19.17 1.64 | 75.42 3.86 | 39.33 3.96 | 44.93 9.10 | 67.64 0.85 | 47.71 6.05 | 14.45 2.39 | 42.04 1.38 | 11.85 0.31 | |
| 16 | 34.28 1.26 | 57.51 0.90 | 40.89 3.72 | 19.29 1.09 | 74.84 3.21 | 39.13 3.66 | 46.24 13.83 | 67.37 1.99 | 48.22 12.17 | 16.33 3.62 | 43.40 1.50 | 11.97 0.30 | |
| 32 | 35.88 1.72 | 56.53 2.03 | 37.08 5.26 | 19.11 1.33 | 73.88 3.54 | 38.94 3.43 | 55.69 3.88 | 69.31 0.71 | 58.58 7.57 | 13.92 4.30 | 44.44 1.93 | 12.51 0.14 | |
| 64 | 37.63 2.06 | 54.28 2.15 | 34.93 2.97 | 18.85 1.37 | 72.54 3.62 | 39.37 4.32 | 46.59 5.49 | 66.95 0.71 | 47.91 5.72 | 19.71 3.10 | 43.82 1.72 | 12.59 0.25 | |
Time-Series Anomaly Detection
Time-Series Anomaly Detection identifies unusual patterns or events in sequential data, from sensor readings and network traffic to health and industrial telemetry.
See all 226 papers