Adaptive tree-reinforced clustering using hierarchical social relations and Q-learning in dynamic data environments.

Efficient clustering in large-scale, heterogeneous, and evolving data systems remains a core challenge in data-driven intelligence. Traditional clustering models often face limitations related to computational complexity, noise vulnerability, and lack of adaptability. To overcome these issues, this study introduces the Adaptive Tree-Reinforced Clustering (ATRC) framework, a hybrid model that fuses the Tree Social Relation (TSR) mechanism with Q-Learning-based reinforcement optimization. In this model, TSR establishes hierarchical connections among data entities to facilitate the rapid creation of preliminary clusters, while Q-Learning dynamically refines cluster assignments through iterative state-action evaluation to avoid local minima. The cooperative interaction between TSR's structural organization and Q-Learning's adaptive exploration enhances both convergence stability and accuracy, particularly in uncertain or noisy conditions. Experimental results on a heterogeneous IoT sensor network with 32 nodes and four adaptive clusters demonstrate that ATRC achieves superior performance over LSCC-RL, CRL-AC, and CRAQL benchmarks yielding up to 13% higher packet delivery rates, 28% lower energy consumption, and nearly 0.9 × faster convergence. These results confirm that integrating reinforcement-based learning with hierarchical modeling yields a scalable and energy-efficient clustering framework suitable for real-time IoT, fog computing, and data-intensive intelligent systems.