Advanced Parametric Methods for Short-Term Traffic Forecasting in the Era of Big Data

Abstract

We live in the era of big data in all fields of activity and intensity. From econometrics and bioinformatics to robotics and aviation and from computational linguistics and social networks to traffic and transportation analytics, big data is the dominating factor of progress. Especially in the field of Intelligent Transportation Systems (ITS), the plethora of multisource traffic data has given a tremendous boost to the development of sophisticated systems for the confrontation of the several traffic related problems. One of the most challenging and at the same time crucial traffic related problems, which has significant impact in many ITS systems (e.g. Advanced Traveler Information Systems, multimodal routing systems, dynamic pricing systems, etc.), is the accurate and real-time traffic forecasting. The task of traffic forecasting, i.e. predicting the state of traffic in large scale urban and inter-urban networks within multiple intervals ahead in time, includes addressing several subproblems, like data acquisition from multiple sources (e.g. inductive loop detectors, moving vehicles, traffic cameras, etc.), preprocessing (outlier detection, missing data imputation, map-matching, etc.), integration and storage, design and development of complex algorithmic methods, overall network coverage of the forecasting results, performance issues, etc. In this chapter, the several state-of-the-art methods used in all aspects of the traffic forecasting problems are presented, with particular emphasis given on both the algorithmic and the efficiency aspects of the problem, in the light of the large amounts of available traffic data. In particular, the design of advanced traffic forecasting algorithms in large scale urban and inter-urban road networks are described along with their implementation and utilization on large amounts of real world traffic data.