Graph embedded rules for explainable predictions in data streams.

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Decision support systems often prioritize accuracy over interpretability.
• Data stream mining presents unique challenges due to high data volume, velocity, and concept drift.
• Existing methods like black-box models lack transparency, while white-box models may lack accuracy.

Purpose of the Study:

• To develop an interpretable machine learning model for data stream classification.
• To address the trade-off between accuracy and interpretability in dynamic data environments.
• To propose a novel approach for extracting understandable rules from data streams.

Main Methods:

• Modeling data streams using attribute space graphs for rule extraction.
• Employing ensemble techniques with graph variants to handle concept drift.
• Updating graph-based models over time to maintain performance.

Main Results:

• The proposed graph-based ensemble method achieved superior classification results.
• Outperformed six existing rule-based algorithms across twelve streaming datasets.
• Demonstrated effective handling of concept drift and improved model accuracy.

Conclusions:

• Graph-based modeling offers a promising solution for interpretable data stream mining.
• The ensemble approach effectively mitigates concept drift and enhances predictive performance.
• This method provides a transparent alternative to traditional black-box and less accurate white-box models.