Traffic Flow Prediction in Sensor-Limited Areas Through Synthetic Sensing and Data Fusion

Traffic flow prediction is an important feature for smart cities, as it helps in implementing effective traffic policies. However, accurate prediction and successful traffic management rely on reliable traffic information, which may not always be available due to the deployment and management costs of a specialized traffic intensity sensor infrastructure. We investigate the possibility of collecting alternative data sources and employing data fusion methodologies to build usable measurements. Hence, in this research, we focus on leveraging artificial intelligence techniques, data fusion, and synthetic sensing, to improve the accuracy of traffic flow prediction in regions with limited sensor infrastructure. The considered alternative measurements are source-destination datasets (e.g., those provided by mobile maps providers), meteorological data, bus trajectory information, and path and delay, obtained from metropolitan transport service providers. By fusing and analyzing these data sources, it becomes possible to predict traffic flow in a specific and localized area. In this research, our focus is specifically on the city of Issy le Moulineax, France. The study has analyzed the fused datasets using three distinct machine learning techniques, long-short term memory (LSTM), Facebook Prophet (FB), and Neural Prophet, to identify the most suitable model. The goal is to help cities have access to accurate traffic flow predictions without the need to deploy a specialized traffic sensor network. These data can then be used to facilitate urban mobility planning.