Akdag, Hakan und Bui, Anh Thu und Hußfeldt, Natalie und Heisenberg, Gernot (2026) A Hybrid Framework for Time Series Anomaly Detection with LLMs. CIKM 26, 2026-11-07 - 2026-11-11, Rom, Italien. doi: 10.1145/3799682.3840786.
|
PDF
1MB |
Kurzfassung
Large Language Models (LLMs) are increasingly explored for time series anomaly detection (TSAD), yet their direct use often remains inferior to methods designed specifically for TSAD. In this work, we study whether this limitation arises primarily from weak forecasting or from insufficient exploitation of forecast residuals. Across LLaMA-3.1-8B-Instruct, Mistral-7B, and DeepSeek-R1-Distill-Qwen-7B, we find that parameter-efficient fine-tuning can improve forecasting accuracy, but does not consistently improve anomaly detection under fixed residual thresholding. The primary driver of detection performance is not forecasting quality, but how the resulting residuals are analyzed and refined in subsequent stages. Motivated by this finding, we propose a Forecaster–Detector–Imputer pipeline that restricts the LLM to forecasting normal temporal behavior, applies classical detectors to the resulting residuals using adaptive windowing and contamination-aware thresholding, and refines candidate detections through imputation via a time-series foundation model followed by a second detection pass. The proposed framework reaches F1 = 0.789 with LLaMA and F1 = 0.785 with Mistral, roughly doubling the zero-shot baselines, with results confirmed on an independent NAB benchmark. These results suggest that LLMs are most effective for TSAD when used as forecasting components within a hybrid residual-analysis pipeline, and that the primary bottleneck lies in post-hoc residual exploitation rather than forecasting quality alone
| elib-URL des Eintrags: | https://elib.dlr.de/229044/ | ||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Dokumentart: | Konferenzbeitrag (Poster) | ||||||||||||||||||||
| Titel: | A Hybrid Framework for Time Series Anomaly Detection with LLMs | ||||||||||||||||||||
| Autoren: |
| ||||||||||||||||||||
| Datum: | 2026 | ||||||||||||||||||||
| Referierte Publikation: | Nein | ||||||||||||||||||||
| Open Access: | Ja | ||||||||||||||||||||
| Gold Open Access: | Nein | ||||||||||||||||||||
| In SCOPUS: | Nein | ||||||||||||||||||||
| In ISI Web of Science: | Nein | ||||||||||||||||||||
| DOI: | 10.1145/3799682.3840786 | ||||||||||||||||||||
| Status: | veröffentlicht | ||||||||||||||||||||
| Stichwörter: | Time Series Anomaly Detection; Large Language Models; Hybrid Models; Residual-Based Detection; Parameter-Efficient Fine-Tuning; Imputation; Few-Shot Forecasting | ||||||||||||||||||||
| Veranstaltungstitel: | CIKM 26 | ||||||||||||||||||||
| Veranstaltungsort: | Rom, Italien | ||||||||||||||||||||
| Veranstaltungsart: | internationale Konferenz | ||||||||||||||||||||
| Veranstaltungsbeginn: | 7 November 2026 | ||||||||||||||||||||
| Veranstaltungsende: | 11 November 2026 | ||||||||||||||||||||
| HGF - Forschungsbereich: | Luftfahrt, Raumfahrt und Verkehr | ||||||||||||||||||||
| HGF - Programm: | Raumfahrt | ||||||||||||||||||||
| HGF - Programmthema: | Technik für Raumfahrtsysteme | ||||||||||||||||||||
| DLR - Schwerpunkt: | Raumfahrt | ||||||||||||||||||||
| DLR - Forschungsgebiet: | R SY - Technik für Raumfahrtsysteme | ||||||||||||||||||||
| DLR - Teilgebiet (Projekt, Vorhaben): | R - CERES | Computing-Effizienz und -Resilienz für Raumfahrt-Software | ||||||||||||||||||||
| Standort: | Köln-Porz | ||||||||||||||||||||
| Institute & Einrichtungen: | Institut für Softwaretechnologie > High-Performance Computing Institut für Softwaretechnologie | ||||||||||||||||||||
| Hinterlegt von: | Akdag, Dr. Hakan | ||||||||||||||||||||
| Hinterlegt am: | 11 Nov 2026 08:31 | ||||||||||||||||||||
| Letzte Änderung: | 11 Nov 2026 08:31 |
Nur für Mitarbeiter des Archivs: Kontrollseite des Eintrags