Tracking Correlations Between Multiple Data Streams Through Evolutionary Regressor Chains.

Area of Science:

• Machine Learning
• Data Science
• Time Series Analysis

Background:

• Real-world data involves multiple simultaneous correlational data streams.
• Nonstationary data streams and evolving correlations pose challenges for machine learning.
• Existing models struggle to adapt to dynamic inter-stream correlations.

Purpose of the Study:

• To develop a novel ensemble model capable of tracking and leveraging dynamic correlations between data streams.
• To enhance the effectiveness of machine learning models in nonstationary environments.
• To address the challenge of changing correlations in real-world data streams.

Main Methods:

• Proposed an ensemble chain-structured model: Evolutionary Regressor Chains (RCs).
• Developed a heuristic order searching approach for optimal chain configuration and dynamic updates.
• Introduced a method to reduce computational complexity while preserving ensemble diversity.
• Established theoretical foundations using dynamic regret analysis for optimal adaptation.

Main Results:

• Evolutionary RCs effectively track the dynamicity of correlations across data streams.
• The heuristic search method successfully updates chains over time.
• The proposed complexity reduction method maintains ensemble diversity.
• Dynamic regret analysis confirms optimal adaptation capabilities.

Conclusions:

• Evolutionary RCs offer a robust solution for machine learning with dynamic correlational data streams.
• The model demonstrates superior performance in environments with nonstationary and evolving correlations.
• The approach provides a computationally efficient and adaptive method for complex data stream analysis.