2019/11/29 by Julien Monteil, Monteil, Julien, Anton Dekusar +7
Engineering · Social Sciences · #FOS: Computer and information sciences #Human Mobility and Location-Based Analysis #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Traffic Prediction and Management Techniques #Transportation Planning and Optimization
paper · pdf · doi:10.48550/arxiv.1911.13042
openalex publication_date 2019/11/29 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
The transport literature is dense regarding short-term traffic predictions,\nup to the scale of 1 hour, yet less dense for long-term traffic predictions.\nThe transport literature is also sparse when it comes to city-scale traffic\npredictions, mainly because of low data availability. In this work, we report\nan effort to investigate whether deep learning models can be useful for the\nlong-term large-scale traffic prediction task, while focusing on the\nscalability of the models. We investigate a city-scale traffic dataset with 14\nweeks of speed observations collected every 15 minutes over 1098 segments in\nthe hypercenter of Los Angeles, California. We look at a variety of\nstate-of-the-art machine learning and deep learning predictors for link-based\npredictions, and investigate how such predictors can scale up to larger areas\nwith clustering, and graph convolutional approaches. We discuss that modelling\ntemporal and spatial features into deep learning predictors can be helpful for\nlong-term predictions, while simpler, not deep learning-based predictors,\nachieve very satisfactory performance for link-based and short-term\nforecasting. The trade-off is discussed not only in terms of prediction\naccuracy vs prediction horizon but also in terms of training time and model\nsizing.\n