A Sustainability-Aware Federated Graph Attention Framework for Supply Chain Process Modelling

Modern supply chains operate as highly interconnected networks characterized by decentralization, data silos, and increasing sustainability constraints. While graph neural networks (GNNs) have demonstrated strong performance in modelling such relational systems, their practical deployment is hindered by limited data sharing across organizations. Federated learning (FL) offers a promising solution by enabling collaborative model training without exposing proprietary data, yet existing approaches rarely integrate graph structure and sustainability objectives simultaneously. This study proposes a sustainability-aware federated graph attention framework for supply chain process modelling. The framework combines Graph Attention Networks (GATs) with federated optimization to learn from decentralized, partially observable supply chain subgraphs while embedding environmental considerations directly into the attention mechanism. A synthetic multi-tier supply chain case study is developed to evaluate the approach under realistic data-governance constraints. Experimental results show that while centralized graph learning achieves the highest predictive accuracy, the proposed sustainability-aware federated GAT attains competitive performance relative to standard federated baselines, while systematically reducing reliance on carbon-intensive transport links. An ablation analysis demonstrates a smooth and controllable trade-off between predictive accuracy and sustainability alignment through a single policy parameter. The findings highlight the feasibility of privacy-preserving, sustainability-informed graph learning for supply chain process modelling and provide a principled foundation for environmentally aligned AI deployment in multi-enterprise settings.