An Integrated Cloud–HPC Framework for Traffic Prediction and Simulation using Mobile Crowdsensing Networks

Urban traffic congestion has emerged as a critical challenge in metropolitan cities worldwide, leading to increased fuel consumption, prolonged travel times, and environmental degradation. This paper proposes a novel integrated Cloud-HPC framework that leverages mobile crowdsensing networks for traffic prediction and simulation. The framework combines deep learning models—specifically Long Short-Term Memory (LSTM) networks and Graph Neural Networks (GNNs)—with GPU-accelerated high-performance computing to enable real-time traffic forecasting. We utilize the METR-LA traffic dataset comprising traffic speed measurements from 207 sensors across Los Angeles County, preprocessed through a cloud-based data pipeline using Apache Spark. Our LSTM model achieves a Root Mean Square Error (RMSE) of 6.31, Mean Absolute Error (MAE) of 4.04, and Mean Absolute Percentage Error (MAPE) of 12.8%, while the GNN model further improves accuracy with RMSE of 4.82, MAE of 3.15, and MAPE of 9.8%. The GPU-accelerated training demonstrates significant ssspeedup, reducing training time from 846.53 minutes on CPU to 86.38 minutes on NVIDIA Tesla T4 GPU—a 9.8x acceleration for LSTM, and from 865.87 minutes to 95.15 minutes—a 9.1x acceleration for GNN. Scalability experiments show near-linear speedup with increasing data volumes, processing 100,000 vehicle trajectories in 6.3 minutes. The SUMO-based traffic simulation framework successfully models four congestion scenarios: accidents, road closures, festival traffic, and increased vehicle density, demonstrating average travel time reductions of 23.4% through predictive rerouting. This integrated approach establishes a foundation for next-generation intelligent transportation systems in smart cities.