Dynamic Structure Embedded Online Multiple-Output Regression for Streaming Data.

Area of Science:

• Machine Learning
• Data Science
• Statistical Modeling

Background:

• Online multiple-output regression is crucial for handling multi-dimensional correlated data streams.
• Existing methods face limitations in dynamically adapting to evolving data characteristics and residual error structures.

Purpose of the Study:

• To propose a novel online multiple-output regression method (MORES) for streaming data.
• To enhance prediction accuracy by dynamically learning regression coefficient and residual error structures.
• To improve computational efficiency for real-time data processing.

Main Methods:

• MORES dynamically learns regression coefficient structures for continuous model refinement.
• It leverages the structure of residual errors to boost prediction accuracy.
• Modified covariance matrices and weighted samples are used to track evolving data streams.
• An efficient optimization algorithm and online eigenvalue decomposition are employed.

Main Results:

• Experiments on synthetic and real-world datasets demonstrate MORES' effectiveness and efficiency.
• MORES achieves processing speeds of over 2,000 instances per second.
• The method significantly outperforms state-of-the-art online learning algorithms in speed.

Conclusions:

• MORES offers a significant advancement in online multiple-output regression for streaming data.
• The method provides a robust and computationally efficient solution for complex data modeling.
• Its ability to dynamically adapt to data streams enhances predictive performance.