Integrating Deep Traffic Prediction and Environmental Impact Assessment Using Noisy Real-World Data in Las Vegas

This thesis introduces an integrated framework for advanced traffic prediction and real-time emission estimation, designed to aid urban planning and environmental monitoring. Utilizing a graph-based transformer model, it predicts traffic conditions across the Las Vegas road network, drawing on spatial and temporal data from a large-scale sensor network. The study significantly expands the dataset from 26 to approximately 900 sensors, enhancing predictive accuracy and regional coverage. Inspired by masking techniques and strategies tailored to incomplete datasets, the model effectively handles real-world, noisy data without relying on resource-intensive imputation. Innovative training approaches enable robust traffic flow predictions despite missing or imperfect data. Additionally, traffic predictions are seamlessly integrated with the EMFAC model to provide high-resolution, real-time emission estimates for pollutants such as CO2, NOx, and PM2.5. This system delivers actionable insights into the environmental impacts of traffic, equipping policymakers and urban planners with tools to mitigate air pollution and optimize traffic management.