Contaminated Multivariate Time-Series Anomaly Detection with Spatio-Temporal Graph Conditional Diffusion Models

TL;DR

This paper introduces TSAD-C, an unsupervised method for time-series anomaly detection that effectively handles noisy training data by decontaminating anomalies, modeling variable dependencies, and accurately scoring anomalies, outperforming existing methods.

Contribution

The paper proposes TSAD-C, a novel end-to-end unsupervised approach for contaminated time-series anomaly detection that does not require abnormality labels during training.

Findings

TSAD-C outperforms existing methods on four datasets.

The decontaminator effectively reduces noise impact.

Modeling long-term dependencies improves detection accuracy.

Abstract

Mainstream unsupervised anomaly detection algorithms often excel in academic datasets, yet their real-world performance is restricted due to the controlled experimental conditions involving clean training data. Addressing the challenge of training with noise, a prevalent issue in practical anomaly detection, is frequently overlooked. In a pioneering endeavor, this study delves into the realm of label-level noise within sensory time-series anomaly detection (TSAD). This paper presents a novel and practical end-to-end unsupervised TSAD when the training data is contaminated with anomalies. The introduced approach, called TSAD-C, is devoid of access to abnormality labels during the training phase. TSAD-C encompasses three core modules: a Decontaminator to rectify anomalies (aka noise) present during training, a Long-range Variable Dependency Modeling module to capture long-term intra- and inter-variable dependencies within the decontaminated data that is considered as a surrogate of the pure normal data, and an Anomaly Scoring module to detect anomalies from all types. Our extensive experiments conducted on four reliable and diverse datasets conclusively demonstrate that TSAD-C surpasses existing methodologies, thus establishing a new state-of-the-art in the TSAD field.