Benchmarking Unsupervised Strategies for Anomaly Detection in Multivariate Time Series

TL;DR

This paper evaluates transformer-based models, especially iTransformer, for detecting anomalies in multivariate time series, analyzing parameter effects, label extraction, training data influence, and comparing models across datasets.

Contribution

It introduces a detailed analysis of applying iTransformer to anomaly detection, including parameter tuning, label extraction, training data impact, and comprehensive model comparison.

Findings

iTransformer's performance varies with window and step sizes

Effective label extraction methods improve anomaly detection accuracy

Training with anomalous data affects model robustness

Abstract

Anomaly detection in multivariate time series is an important problem across various fields such as healthcare, financial services, manufacturing or physics detector monitoring. Accurately identifying when unexpected errors or faults occur is essential, yet challenging, due to the unknown nature of anomalies and the complex interdependencies between time series dimensions. In this paper, we investigate transformer-based approaches for time series anomaly detection, focusing on the recently proposed iTransformer architecture. Our contributions are fourfold: (i) we explore the application of the iTransformer to time series anomaly detection, and analyse the influence of key parameters such as window size, step size, and model dimensions on performance; (ii) we examine methods for extracting anomaly labels from multidimensional anomaly scores and discuss appropriate evaluation metrics for such labels; (iii) we study the impact of anomalous data present during training and assess the effectiveness of alternative loss functions in mitigating their influence; and (iv) we present a comprehensive comparison of several transformer-based models across a diverse set of datasets for time series anomaly detection.