In a Streaming World, Should You Stand Still? A Comprehensive Benchmark of Anomaly Detection in Streams

Magali Parrino, Antoine Ajenjo, Emmanuel Remy, Pierre Stephan, Pierre Senellart, Paul Boniol

arXiv: 2609.39215Published: September 30, 2026

AI summary

TL;DR

This paper compares streaming and static time series anomaly detection methods, finding that static methods outperform streaming approaches in most settings. The study highlights a gap between existing streaming methods and modern TSAD requirements. Static methods are preferred due to their accuracy and efficiency.

Why It Matters

The findings have significant implications for real-world applications, such as monitoring systems and predictive maintenance, where accurate anomaly detection is crucial. The results can inform the development of more effective TSAD systems. This research can lead to improved performance and efficiency in various industries.

Key Findings

static TSAD methods outperform streaming approaches!.
streaming methods are not advantageous in realistic TSAD scenarios
distribution-drift dataset TSB-drift is proposed

Topics

time series analysis

anomaly detection

streaming data

machine learning

evaluation metrics

Thank you to arXiv for use of its open access interoperability. arxiv.org. aibrief.ai is not affiliated with or endorsed by arXiv.