Benchmarking Unsupervised Strategies for Anomaly Detection in Multivariate Time Series

2025-06-25

AI Summary

Multivariate time-series anomaly detection is highly challenging due to unknown anomaly patterns and strong inter-variable dependencies. This work systematically investigates the application of iTransformer to unsupervised multivariate anomaly detection, proposing a novel anomaly scoring framework that aggregates multidimensional reconstruction errors and employs adaptive thresholding. We quantitatively analyze the impact of key hyperparameters—including window size, stride, and embedding dimension—and assess model robustness under contamination of training data with latent anomalies. Extensive experiments are conducted across multiple real-world benchmarks, comparing iTransformer against state-of-the-art Transformer variants. Results demonstrate that iTransformer achieves statistically significant improvements in F1-score and point-adjusted F1, validating its superior capability in capturing dynamic cross-variable dependencies. Moreover, this study provides a reproducible, empirically grounded guideline for model selection, hyperparameter configuration, and evaluation practice in unsupervised anomaly detection.

Technology Category

• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Machine Learning: Calibration & Uncertainty Quantification
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

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.

Problem

• Exploring transformer-based methods for multivariate time series anomaly detection
• Analyzing parameter impacts on anomaly detection model performance
• Evaluating strategies to handle anomalous data during model training

Innovation

• iTransformer for anomaly detection
• Analyzing key parameters impact
• Comparing transformer-based models comprehensively