City-Wide Traffic Prediction using Aggregated Mobile Network Activity Data

Accurate city-wide traffic prediction is essential for intelligent mobility management, urban planning, and traffic operations. Conventional approaches often depend on fixed traffic sensors or individual vehicle trajectories, which may provide uneven spatial coverage. This paper presents a framework based on aggregated cellular network activity as a large-scale source of mobility information for predicting road travel time and speed. The source research introduces two complementary deep-learning models: the Network Insight Model, which uses aggregated cellular activity alone, and the Integrated Mobility Model, which combines cellular activity with Bluetooth sensor observations. Cellular activity is represented through Network Activity Tensors (NAT), while Bluetooth observations are represented through Bluetooth Graph Tensors (BGT). Graph Attention Networks (GAT), Gated Recurrent Units (GRU), and convolutional layers are used to learn spatial and temporal relationships and generate city-wide origin–destination predictions. The models are evaluated using real-world data from York Region, Canada, with MAE, RMSE, R², MAPE, and Travel Time Index (TTI). The reported results show that combining cellular and Bluetooth data improves both travel-time and speed prediction compared with the cellular-only model and the selected baselines.