A Two-Stage Model for Real-time Taxi Demand Prediction Using Data from the Web

Timely and accurate anticipation of phenomena in a variety of applications, such as economics, energy and transportation, is considered necessary for their smooth operation. Up to the present time, several efforts have been made in the transport sector to exploit information related to supply and demand for the formulation of accurate forecasting models. However, it is widely known that the model that could provide us with very accurate demand predictions under any circumstances has not yet been formed. Many factors in daily life, such as special events and traffic disruptions, overturn traffic system's balance, and the reliability of forecast models decreases significantly. The main focus of this research is the analysis, evaluation, and forecasting of prediction model’s residuals in a real-time taxi demand forecasting framework. The authors comprise a deep learning architecture that is based on Fully-Connected dense layers. Publicly available taxi data from New York are explored, as well as semantic information combinations, that are typically neglected from modern techniques. The analysis focuses on two main areas, where significant fluctuations in demand are observed, due to popular venues located in the area. The performance of the proposed two-stage process with the inclusion of residuals’ forecasts, is improved considerably.