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

In distributed data environments, classification tasks are challenged by inconsistencies across independently maintained sources. These environments are inherently characterized by high informational uncertainty. Our framework addresses this challenge through a structured process designed for the reduction of entropy in the overall decision-making process. This paper proposes a novel framework that integrates conflict analysis, coalition formation, decision tree induction, and decision template fusion to address these challenges. The method begins by identifying compatible data sources using Pawlak's conflict model, forming coalitions that aggregate complementary information. Each coalition trains a decision tree classifier, and the final decision is derived through decision templates that fuse probabilistic outputs from all models. The proposed approach is compared with a variant that does not use coalitions, where each local source is modeled independently. Additionally, the framework extends previous work based on decision rules by introducing decision trees, which offer greater modeling flexibility while preserving interpretability. Experimental results on benchmark datasets from the UCI repository demonstrate that the proposed method consistently outperforms both the non-coalition variant and the rule-based version, particularly under moderate data dispersion. The key contributions of this work include the integration of coalition-based modeling with decision trees, the use of decision templates for interpretable fusion, and the demonstration of improved classification performance across diverse scenarios.