Pre-trained Time Series Foundation Models for Traffic Flow Prediction Using Open Government Data: The case of Attica traffic OGD

Rapid urbanization and increase in vehicular traffic in metropolitan areas necessitate advanced predictive models to enhance traffic management systems.Open Government Data (OGD) portals provide accessible, high-quality traffic datasets, offering a valuable resource for developing and evaluating such models.This paper investigates the effectiveness of pre-trained Time Series Foundation Models, specifically Lag-Llama, in forecasting urban traffic flow using open government traffic data from the Greek OGD portal.We assess Lag-Llama's performance in both zero-shot and fine-tuned settings and compare it with traditional deep learning models, including Long Short-Term Memory (LSTM) networks and Graph Neural Networks (GNNs).Our experimental results demonstrate that Lag-Llama, especially when fine-tuned on traffic OGD, outperforms conventional models in prediction accuracy.These findings highlight the potential of combining foundation models with traffic open government data to advance intelligent transportation systems and contribute to the development of more robust and reliable traffic forecasting models, as well as provide innovative and efficient public services for citizens and businesses.