Graph-augmented transformer networks and explainable AI for economic impact forecasting in disrupted supply chains.

This study proposes a hybrid AI model which integrates Graph Neural Networks and Transformer models to predict the economic consequence of the disruption of the supply chains with unprecedented accuracy (MAPE: 3.7%). The framework is based on multi-dimensional data from 137 companies in 23 countries, and shows the non-linear cascading effects as well as the difference between output and throughput effects. There's a dynamic resilience scoring system (91.4% accuracy) and customized explainability techniques (attributional, counterfactual and strategic) for actionable insights. Empirical tests confirm that the prediction of the impacts of the earthquakes in the long term is improved by 27.3% compared to the traditional econometric model and the ML model. The work takes a theoretical approach to resilience and connects it with tangible economic results, providing relevant policy makers with tools to reduce risks and increase stability. Its contributions encompass an economic consequences layer to enable economic interpretation, a propagation model for disruption effects, and a scalable architecture for global supply chains. The framework's effectiveness at industry and geographical levels highlights its relevance for strategic and operational decision making.