Robust Group Anomaly Detection for Quasi-Periodic Network Time Series

2022-07-01

IEEE Transactions on Network Science and Engineering

AI Summary

This work addresses the challenge of group-level quasi-periodic anomaly detection in multivariate time series from sensor networks—characterized by variable periods, asynchrony, and unequal lengths. We propose seq2GMM, a novel framework that maps raw sequences to Gaussian Mixture Model (GMM) parameters, enabling robust representation learning and anomaly scoring at the population level. To optimize GMM fitting efficiently, we design a surrogate-model-based optimization algorithm with rigorous convergence guarantees. The method unifies sequence embedding, probabilistic GMM modeling, and interpretable diagnostics: it achieves precise anomaly identification while supporting expert interpretation via semantic inversion of GMM parameters (e.g., component means and covariances reflect characteristic patterns and deviations). Evaluated on multiple public benchmark datasets, seq2GMM outperforms state-of-the-art methods, delivering an average 6.2% improvement in F1-score and strong interpretability—bridging high detection accuracy with actionable diagnostic insights.

Technology Category

Data Mining & Knowledge Management: Anomaly/Outlier DetectionMachine Learning: Time-Series/Data StreamsReasoning under Uncertainty: Relational Probabilistic Models

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 graphsSemantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semanticsSearch and Retrieval-Augmented AI: Web query analysis, representation and understanding

Abstract

Many real-world multivariate time series are collected from a network of physical objects embedded with software, electronics, and sensors. The quasi-periodic signals generated by these objects often follow a similar repetitive and periodic pattern, but have variations in the period, and come in different lengths caused by timing (synchronization) errors. Given a multitude of such quasi-periodic time series, can we build machine learning models to identify those time series that behave differently from the majority of the observations? In addition, can the models help human experts to understand how the decision was made? We propose a sequence to Gaussian Mixture Model (seq2GMM) framework. The overarching goal of this framework is to identify unusual and interesting time series within a network time series database. We further develop a surrogate-based optimization algorithm that can efficiently train the seq2GMM model. Seq2GMM exhibits strong empirical performance on a plurality of public benchmark datasets, outperforming state-of-the-art anomaly detection techniques by a significant margin. We also theoretically analyze the convergence property of the proposed training algorithm and provide numerical results to substantiate our theoretical claims.

Problem

• Detect anomalous quasi-periodic network time series
• Explain model decisions for human experts
• Train efficient anomaly detection models

Innovation

• Sequence to Gaussian Mixture Model framework
• Surrogate-based optimization algorithm
• Theoretical convergence analysis