Minimal dataset for "Predictive analytics for supply chain resilience in urban infrastructure networks using graph convolutional networks" (PONE-D-26-10190R1)

This dataset supports the manuscript "Predictive analytics for supply chain resilience in urban infrastructure networks using graph convolutional networks" (submitted to PLOS ONE, 2026). The study proposes a three-stage analytical framework integrating complex network analysis, Graph Convolutional Network (GCN)-based flow prediction, and network resilience stress testing, validated on a New York City urban road network dataset. The GCN model achieves an RMSE of 5.8 and R² of 0.92 for supply chain flow prediction, outperforming ARIMA, SVR, LSTM, and XGBoost baselines. Resilience analysis identifies a crossover effect at approximately 30% node removal where random failures surpass targeted attacks in cumulative connectivity loss. Dataset contents: supply_chain_resilience_v2.db — Cleaned SQLite database (six tables): road network topology (6,302 nodes, 13,801 edges) derived from OpenStreetMap; 96,641 vehicle collision records from NYC Open Data (Oct 2024–Dec 2025); 100,000 automated traffic volume counts from NYC DOT; and two analysis-ready derived tables filtered to the study period (Oct 2024–Jun 2025). table1_dataset_overview.csv through table9_statistical_comparison.csv — Tabular data underlying all nine manuscript tables (network metrics, model performance, ablation study, resilience test results, and statistical significance tests). data_collection.py — Script to download raw data from public APIs (OpenStreetMap via OSMnx; NYC Open Data). plot_figures.py — Script to reproduce all ten manuscript figures from the database. Data sources: OpenStreetMap (ODbL v1.0); NYC Open Data Motor Vehicle Collisions–Crashes (h9gi-nx95); NYC DOT Automated Traffic Volume Counts (7ym2-wayt). All data are publicly available and free of personally identifiable information.