Conditional normalizing flow for multivariate time series anomaly detection.

Area of Science:

• Data Science
• Machine Learning
• Time Series Analysis

Background:

• Multivariate time series data is prevalent across critical sectors like healthcare and industry.
• Detecting anomalies in this data is difficult due to its high dimensionality, temporal dependencies, and lack of labeled examples.

Purpose of the Study:

• To develop an unsupervised algorithm for accurate anomaly detection in multivariate time series.
• To address the challenges of high dimensionality, temporal complexity, and label scarcity in time series data.

Main Methods:

• Proposed an Attention Factorization Normalizing Flow (AFNF) algorithm.
• Utilized time series factorization and an attention mechanism to model conditional densities.
• Incorporated adjacency contrasting and global location encoding to capture temporal dynamics.

Main Results:

• AFNF demonstrated superior performance in anomaly detection across three real-world datasets.
• The method effectively estimated data distributions and identified anomalies.
• Achieved state-of-the-art results in unsupervised multivariate time series anomaly detection.

Conclusions:

• The proposed AFNF algorithm is effective for unsupervised anomaly detection in multivariate time series.
• The integration of attention, factorization, and normalizing flows provides a robust approach.
• The method shows significant potential for applications in server monitoring, industrial processes, and healthcare.