PROJECT TITLE :

Deep Learning for Road Traffic Forecasting Does it Make a Difference

ABSTRACT:

It has been demonstrated that the methods of deep learning are flexible enough to model complex phenomena. This has also been the case with Intelligent Transportation Systems, in which a number of subfields, such as vehicular perception and traffic analysis, have broadly embraced deep learning as a fundamental modeling technology. The capability of Deep Learning to deliver good results has generated a prevalent inertia towards using Deep Learning models, particularly in short-term traffic forecasting. This is the case despite the fact that examining in depth the benefits and drawbacks of these models would be beneficial. In this particular research area of Intelligent Transportation Systems, the primary objective of this paper is to conduct an in-depth evaluation and assessment of the current state of the art regarding the application of Deep Learning. In order to accomplish this, we expand on the findings that were derived from a study of publications that were produced in the most recent years, based on two taxonomic criteria. An after-the-fact critical analysis is being carried out in order to formulate questions and kick off a necessary discussion regarding the problems associated with using Deep Learning for traffic forecasting. The research is finished off with a comparison of various short-term traffic forecasting methods applied to traffic datasets of varying types, with the goal of covering a wide range of potential scenarios. The results of our experiments indicate that Deep Learning might not be the most effective modeling method in all circumstances. This reveals a few caveats that have not been taken into consideration up to this point and should be addressed by the community in prospective studies. These new insights bring to light new challenges and opportunities for research in the field of road traffic forecasting. These new challenges and opportunities are enumerated and thoroughly discussed with the goal of motivating and guiding future research efforts in this field.