Detecting the Undetectable: Enhancing Unsupervised Time Series Anomaly Detection via Active Learning

Jul 16, 2026

Upload by Anda

Despite advances in industrial AI, reliable detection of subtle, noisy anomalies in complex time series remains unresolved. Labeling is often prohibitively costly in large-scale deployments, so unsupervised learning is widely used. Existing unsupervised methods struggle to separate near-normal anomalies from normal behavior and remain sensitive to noise inside normal samples. This work proposes an active-learning framework with masked reconstruction feedback and minimax robustness, improving AUC by 12.39% across 28 testcases and seven backbone models.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages34

Reading Time1 h 26 min

Introduction

Motivation for unsupervised anomaly detection

Limitations: noise sensitivity and near-normal anomalies

Proposed approach: active learning framework

arXiv :2607 .00720v 1 [ cs .LG] 1 Jul 2026 Detecting the Undetectable: Enhancing Unsupervised time series Anomaly Detection via Active Learning Seung Hun Hana , Hyeongwon Kangb , Jinwoo Parkc , Pilsung Kangc,∗ aLG CNS, 71 Magokjungang 8-ro, Gangseo-gu, Seoul, Republic of Korea b Department of Industrial & Management Engineering, Korea University, 126-16 Anam-dong 5-ga, Seongbuk-gu, Seoul, Republic of Korea c Department of Industrial Engineering, Seoul National University, Gwanak-ro 1, Gwanak-gu, Seoul, Republic of Korea Abstract Despite the increasing sophistication of industrial AI systems, the ability to reliably detect subtle and noisy anomalies in complex time series data remains a critical yet unresolved challenge. In large-scale industrial applications, labeling time series data is often prohibitively expensive and time-consuming, making unsupervised learning a practical and widely adopted approach. However, existing unsupervised methods frequently struggle to distinguish nearnormal anomalies from normal patterns and are vulnerable to noise contamination within normal samples. To address these limitations, we propose a novel framework that leverages active learning to iteratively enhance the performance of unsupervised models. Our framework’s core contributions are (1) a masked time-series reconstruction feedback strategy that forces the model to learn robust temporal dependencies, and (2) a minimax learning strategy that promotes robustness by differentially treating normal and abnormal samples. This process encourages the model to better capture the dynamics of subtle and noisy patterns. The proposed framework is evaluated across 28 testcases involving four multivariate time-series datasets and seven unsupervised backbone models. Experimental results demonstrate a 12 .39% improvement in AUC compared to the original models, confirming that our method can be readily integrated into existing unsupervised reconstruction-based anomaly ∗ Corresponding author Email addresses: [seunghun.han@lgcns.com](seunghun.han@lgcns.com) (Seung Hun Han), [hyeongwon_kang@korea.ac.kr](hyeongwon_kang@korea.ac.kr) (Hyeongwon Kang), [jinwoo_park@snu.ac.kr](jinwoo_park@snu.ac.kr) (Jinwoo Park), [pilsung_kang@snu.ac.kr](pilsung_kang@snu.ac.kr) (Pilsung Kang) detection systems to significantly enhance their performance. Keywords: Multivariate time series, Unsupervised time series anomaly detection, Active learning 1. Introduction In today’s increasingly automated and sensor-rich industrial environments, vast streams of time series data are continuously generated, capturing the dynamic behaviors of machines, processes, and entire systems. Amid this data deluge, even minor anomalies can signal critical issues such as equipment failure, cyberattacks, or financial risk, underscoring the importance of timely and accurate detection. time series anomaly detection, which aims to identify patterns that deviate significantly from normal behavior, has therefore become a cornerstone of operational reliability and industrial risk mitigation [1] . As modern systems grow more complex and interconnected, the need for intelligent and scalable anomaly detection tools has become more urgent than ever [1, 2] . This growing demand has spurred intense research activity focused on developing effective methods for monitoring and detecting anomalies in multivariate time series data [1, 3, 4] . Recent research on time series anomaly detection has increasingly emphasized unsupervised learning methods, primarily due to the practical challenges of obtaining high-quality labeled data in real-world applications [2, 5, 6] . Annotating time series data is not only labor-intensive and timeconsuming, but also requires substantial domain expertise and a nuanced understanding of temporal dependencies and inter-variable relationships [7, 8, 9] . As a result, most studies have focused on modeling normal behavioral patterns without labeled anomalies, identifyin

FAQs

Why is unsupervised time series anomaly detection widely used in industrial settings?

Because labeling time series data is often prohibitively expensive and time-consuming. Unsupervised learning can detect deviations without requiring labeled anomalies.

What limitations do existing unsupervised methods face?

They can misclassify noisy normal fluctuations as anomalies, causing high false positives. They may also miss subtle anomalies that mimic normal patterns, leading to false negatives.

How does the proposed framework improve unsupervised anomaly detection?

It uses active learning with masked time-series reconstruction feedback to learn robust temporal dependencies and a minimax learning strategy that differentially treats normal versus abnormal samples, improving robustness and AUC.

Jul 16, 2026

Upload by Anda