Distributed Data Classification with Coalition-Based Decision Trees and Decision Template Fusion

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• Distributed data environments present classification challenges due to inconsistencies and high informational uncertainty across independent sources.
• Existing methods struggle with data dispersion and maintaining interpretability in complex, multi-source datasets.

Purpose of the Study:

• To propose a novel framework for distributed data classification that reduces entropy and improves decision-making accuracy.
• To integrate conflict analysis, coalition formation, decision tree induction, and decision template fusion for robust classification.

Main Methods:

• Utilized Pawlak's conflict model to identify compatible data sources and form coalitions for aggregating complementary information.
• Developed decision tree classifiers within each coalition and employed decision templates for fusing probabilistic outputs from all models.
• Introduced decision trees for enhanced modeling flexibility and interpretability compared to traditional decision rules.

Main Results:

• The proposed coalition-based framework with decision trees consistently outperformed a non-coalition variant and a rule-based approach.
• Performance improvements were particularly notable under moderate data dispersion scenarios.
• Demonstrated enhanced classification accuracy and robustness across diverse benchmark datasets from the UCI repository.

Conclusions:

• The integration of coalition-based modeling with decision trees offers a significant advancement in distributed data classification.
• Decision templates provide an interpretable mechanism for fusing information from multiple data sources.
• The framework effectively addresses informational uncertainty and improves classification performance in distributed environments.