On-Board-Unit big data analytics: from data architecture to traffic forecasting

Accurate traffic information is essential to reduce the adverse effects of road transport in cities. Today, urban planners and policy-makers demand Intelligent Transportation Systems (ITS) to conduct proactive traffic management. In this context, the thesis advances the State-Of-The-Art (SOTA) literature of road Traffic Forecasting (TF) for ITS systems. The work investigates the use of On-Board-Unit (OBU) data related to Heavy-Good Vehicles (HGV) in Belgium. The research addresses three main aspects: the setting up of a lambda architecture to collect, store, and process the OBU data; the exploratory OBU data analysis to understand the value in the context of mobility; advanced analytics to perform TF at road network scale.In the first part, we develop the core components of big data architecture to overcome the limitations of traditional solutions. The platform prototype presents promising results in terms of scalability, fault-tolerance, low latency, and debuggability thanks to the use of big data tools now mature in the field of ITS.In the second part, we strive to identify strategic areas in freight traffic. The density-based techniques applied to the OBU data provide insights into traffic dynamics in the Brussels region.In the final part, we perform advanced predictive analytics. We investigate the use of different strategies to tackle short-term traffic prediction at the urban networks. Then, we apply recent Deep Learning (DL) techniques to perform long-term traffic prediction for the urban and freeway road network. Finally, we propose an original Multi-Task Learning (MTL) model to jointly learn traffic flow and speed. Overall the analysis highlights the superiority of Machine Learning (ML) models against traditional approaches when predicting traffic conditions at the transportation network scale.The proposed thesis shows how the use of OBU data can be beneficial for ITS. In particular, it is possible to understand transport dynamics by processing such data with new open source technologies and advanced ML techniques. The presented work provides traffic management operators and other stakeholders with the basic design principles and tools to analyze and predict freight traffic. The findings suggest organizations involved in the transport sector should pursue new ways to collect, process, merge and analyze traffic data when implementing ITS. In particular, the work highlights how the openness of data, technology, and code are fundamental aspects of trustworthy ITS design. In fact, ITS based on accessible resources leads to more informed and reliable decisions that ameliorate road traffic in cities. Transparent ITS would have the ultimate effect of contributing to an improvement in the mobility and lives of citizens.Doctorat en Sciences