Deep Learning-Based Congestion Forecasting: A Literature Review and Future

The quick improvement of transportation systems gives rise to critical issues, the foremost vital of which is traffic congestion, which has numerous negative impacts such as long time travel and road rage. There are other long-term negative impacts. Forecasting traffic congestion has subsequently gotten to be a key objective in optimising traffic flow and imporving the quality of life for people in cities. Machine learning may be a awesome way to predict traffic flow, but Deep learning techniques have been shown to be more effective in reducing road congestion. The reason of the paper is to conduct a systematic mapping study to examine and categorise studies on deep learning strategies to forecast traffic congestion. Selected articles were categorized and analyzed by channel and year of publication, type of study, research context, type of vehicle and road and deep learning techniques applied to forecast traffic congestion. To deal with this situation, the majority of papers use the classification, prediction, regression techniques. It has also been found that in most studies these algorithms are deployed with the dataset of traffic speed and traffic flow. Many of these deep learning techniques follow a supervised learning, unsupervised learning or a hybrid learning to forecast the preferred data such as Convolutional Neural Networks and Long Short-Term Memory.

I. Introduction

With the urban development of recent years, we have found ourselves faced with problems such as environmental impacts, land consumption, energy consumption, high maintenance cost, accident and safety and traffic congestion in [1]. This unprecedented urban growth is resulting in rapid changes to the transport network in the cities, which is causing significant problems. This urban explosion is also due to the increase of transportations modes, which will also be taken into account in our study.