LLM-Grounded Explainability for Port Congestion Prediction via Temporal Graph Attention Networks

ABSTRACT

Predictive models in logistics often act as black boxes, making them difficult for supply chain managers to trust. This paper integrates Temporal Graph Attention Networks with an LLM-driven reasoning engine to predict port congestion. The LLM translates the complex topological outputs of the graph network into grounded, natural language explanations, bridging the gap between raw algorithmic forecasting and human decision-making.