Overcoming feature scarcity in complex system prediction: An alternative delay embedding.

Area of Science:

• Complex Systems Science
• Machine Learning
• Time Series Analysis

Background:

• Predicting complex system dynamics is challenging due to reliance on target-related features.
• Traditional methods struggle when reliable features are scarce or elusive.
• Existing approaches often require extensive feature engineering.

Purpose of the Study:

• Introduce a novel framework, Alternative Delay Embedding (ADE), for enhanced prediction.
• Develop a method that leverages sequential data without needing explicit target-related features.
• Improve the robustness of predictive models, especially with limited data.

Main Methods:

• Integrate delay embedding with Gaussian process regression.
• Utilize the target's sequential information to generate predictor reconstructions.
• Apply the framework to benchmark dynamical systems and real-world datasets.

Main Results:

• ADE demonstrated robust predictive performance across various model systems (logistic map, Mackey-Glass, Lorenz).
• Validated on diverse real-world data, including sea surface temperature, physiological signals, and financial data.
• Showed enhanced robustness compared to classic methods, particularly with short input sequences.

Conclusions:

• Alternative Delay Embedding (ADE) offers a powerful alternative for time series prediction.
• ADE is particularly valuable when target-related features are difficult to identify or obtain.
• The framework complements existing methods, expanding predictive capabilities in complex systems science.