A Visual Analytics Approach for Hardware System Monitoring with Streaming Functional Data Analysis

Area of Science:

• Data Science
• Statistics
• Time Series Analysis

Background:

• Functional data analysis (FDA) offers benefits for time-dependent phenomena but faces computational challenges with high-dimensional, continuous data.
• Analyzing continuously arriving data requires efficient methods for updating FDA results and identifying anomalies.
• Existing FDA methods can be computationally intensive, especially for long time series and real-time applications.

Purpose of the Study:

• To develop a visual analytics approach for monitoring and reviewing streaming time series data.
• To identify outliers in functional time series data using FDA.
• To address the computational challenges of FDA for high-dimensional, continuously arriving data.

Main Methods:

• Introduction of new incremental and progressive algorithms for performing FDA on streaming data.
• Development of the magnitude-shape (MS) plot to visualize functional magnitude and shape outlyingness.
• Integration of MS plots with an FDA version of principal component analysis for enhanced outlier investigation.

Main Results:

• The proposed algorithms efficiently generate MS plots for timely outlier detection in streaming data.
• The combined use of MS plots and FDA-principal component analysis improves the ability to investigate identified outliers.
• Demonstrated effectiveness through two use scenarios with real-world datasets.

Conclusions:

• The visual analytics approach effectively addresses computational challenges in FDA for streaming time series.
• The MS plot and integrated FDA-PCA provide a powerful tool for identifying and analyzing outliers in functional data.
• The tool's utility was validated by industry experts using real-world streaming datasets.