Traffic Flow Prediction Using Uber Movement Data

The smart city paradigm is closely related to the orderly and sustainable use of the services it provides, on the efficiency of interconnections and communications that take place in an urban context. In this regard, one of the biggest challenges for smart city development relates to the prediction of traffic conditions. In fact, the city's road system has a decisive impact on air pollution, the management of public events, and in general on the efficiency of services offered to people, and thus strongly affects the city's economic development. In recent years, the development of increasingly effective machine learning and deep learning techniques has made a significant contribution to the definition of predictive models in the smart city domain. Deep learning techniques provide efficient results, but need significant computational resources to deal with huge and constantly updating datasets. Very often, however, the traffic data provided by cities are incomplete and insufficient to implement effective deep-learning models. In this paper, a novel solution for defining predictive models of traffic conditions is presented, based on road segmentation and urban traffic-related data, with the aim of dealing with the inherent complexity of geographical datasets. The obtained model has an average accuracy of 94.8%. The proposed architecture is able to reduce the inherent complexity of traffic related data, is easily scalable, can be quickly applied to any urban context.