Integrating Graph Neural Networks with Visualizations for Inter-County Food Trade Prediction

The AI Institute for Intelligent Cyberinfrastructure with Computational Learning in the Environment (ICICLE) aims to democratize artificial intelligence by building scalable tools to advance AI-driven discovery and integration across society. Within the framework provided in the ICICLE AI Training Catalog, I present a graph-based high-performance approach to predict and visualize inter-county food flows across the United States. Using the Expanse supercomputer at SDSC, I implemented ICICLE’s Graph Neural Network (GNN) model on Jupyter Notebooks to process large‑scale datasets of economic, demographic, and geographic features and generate predictions for food trade flows between Freight Analysis Framework (FAF) zones. To address the sparsity of trade networks, a hurdle modeling pipeline is employed, combining a classification stage to predict trade occurrence with a regression stage to estimate trade volume. Beyond prediction, I integrated static visualizations (Matplotlib, NetworkX) and a dynamic, interactive dashboard (Panel, GeoViews, hvPlot) to enable real‑time exploration of trade networks. This combination of machine learning and visualization provides policymakers and researchers with a scalable framework for interpreting complex food trade patterns, supporting data‑driven decisions in food security, infrastructure planning, and economic resilience.