A Survey on Modern Deep Learning Techniques for Traffic Volume Prediction

Traffic volume prediction plays a crucial role in various aspects of transportation management and urban planning, such as traffic light control, traffic flow regulation. Research in this area dates back to the 1970s, mainly exploring statistical models and traditional machine learning techniques. Recently, modern deep learning approaches, including recurrent neural networks, temporal convolutional networks, attention mechanisms, and graph neural networks, have been explored and seen great success. Indeed, compared with previous methods, modern deep learning methods stand out and offer significantly improved prediction accuracy. While there are many surveys, few specifically focus on recent advancements in traffic volume prediction. In this article, we conduct a survey to systematically summarize the modern deep learning technologies for traffic volume prediction. We present a unified taxonomy covering temporal, spatial, complementary, and hybrid designs. We analyze data components and datasets from temporal, spatial, and supplementary perspectives. Moreover, we synthesize experimental results on the PEMS and LargeST benchmarks, with efficiency and sensitivity analyses. Finally, we identify future directions in exploring modern deep learning methods for traffic volume prediction.