Efficient urban monitoring requires methods able to estimate traffic flows on road segments with limited or no sensor coverage. This study validates a traffic flow-scaling model that combines fixed sensors and Floating Car Data in Bologna across 11 sections. Tree-based Machine Learning models, particularly XGBoost, demonstrated excellent performance after hyperparameter optimization. Using the scaled data, a short-term forecasting task (+1h and +4h) showed that tree-based models outperform a neural network based Long Short-Term Memory model in terms of accuracy and robustness. Results indicate that the integration of floating car data and fixed sensors enables accurate spatio-temporal traffic estimation, providing a scalable and cost-effective strategy for urban mobility planning and proactive traffic management.

Rapid urbanization and the growth of cities worldwide have led to increasing traffic congestion, deteriorating air quality, and heightened safety concerns for both drivers and pedestrians. At the same time, advances in technology have provided traffic management organizations with more sophisticated tools to plan and manage urban transport networks. However, deploying sensors across every road in a metropolitan area remains currently cost prohibitive. To address this challenge, this paper proposes a two-stage approach to modelling traffic on roads without sensors. In the first phase, detailed in this study, Floating Car Data is used to estimate traffic flows on sensor-less roads. By employing Machine Learning techniques, the Floating Car Data is scaled to approximate actual traffic volumes and flows along these routes. The second phase builds upon these estimates to forecast future traffic conditions.

The rapid urbanization of our world has led to an ever-increasing urban population, resulting in significant challenges in managing transportation systems. Traffic congestion, with its associated environmental pollution, safety hazards, and prolonged travel times, continues to plague urban areas. Despite numerous efforts to mitigate these issues, the problem persists and hampers urban development. This paper addresses the pressing need for effective traffic flow forecasting of the city of Catania as a critical element in traffic management. Catania’s rapid urbanization has created a complex transportation network as the city’s population has expanded into suburbs and neighbouring areas, causing congestion, pollution, and increased individual motorization due to the high demand for mobility. This has led to severe traffic congestion in the central district, necessitating an efficient traffic management strategy, particularly through the utilization of traffic flow forecasting. A central challenge in traffic flow prediction is the cost and practicality of sensors deployment on every road. To address this issue, the authors propose a novel two-level machine learning approach. The first level employs an unsupervised clustering model to extract patterns from big data generated by sensors, while the second level employs supervised machine learning models for traffic flow forecasting within each cluster. Importantly, this approach enables predictions for roads lacking sensor data by utilizing a really small subset of these new data from alternative sources and assigning roads to appropriate clusters.