Geospatial AI Applications for Reducing Traffic Congestion and Guiding Planning Decisions

ABSTRACT

Traffic congestion remains a major challenge in metropolitan areas, generating substantial mobility, environmental, and economic costs. This study develops a Geospatial Artificial Intelligence (GeoAI) framework for long‐term traffic forecasting, congestion hotspot detection, and planner‐oriented decision support in New York City. Using 15 years of traffic data, the framework integrates ARIMA and LSTM forecasting, H3‐based spatial analysis, clustering, and a customized LLaMA‐based query portal, grounded in a project‐specific traffic knowledge base and used in a zero‐shot configuration that enables planners to explore model outputs through natural‐language interaction. On the 2021–2024 test set, the LSTM model achieved an RMSE of 342.56 vehicles/day, outperforming ARIMA (417.62 vehicles/day) and reducing prediction error by approximately 18%. Forecasts indicate that average daily traffic volume may increase from 12,540 vehicles in 2025 to 19,680 in 2029. Overall, the study demonstrates how GeoAI can support more proactive, spatially explicit, and accessible urban traffic planning.