Traffic level of service prediction by support vector machine, deep neural network and long short-term memory models

Document Type : Research Article


1 Department of civil and environmental engineering, Tarbiat Modares University, Tehran, Iran

2 Associate Professor at the Faculty of Civil and Environmental Engineering

3 Associate Professor at the Faculty of Electrical & Computer Engineering


Short term prediction of traffic parameters and informing them to travelers and transportation operators is a useful tool of advanced traveler information systems. Also, as an advanced traffic management system, it helps to make or maintains the balance between travel demand and supply for the near future. This paper predicts the hourly traffic level of service, which has easily understandable information for all users. Data used in this study is related to 5 sections of a critical suburban road in the north of Iran. This data was collected for five years, and due to its high volume, it considered big data. Long short term memory and deep neural network as two deep learning algorithms and support vector machine as a well-known classifier are trained by the first four years records. Results show that in average long short term memory predictions are more accurate for all sections, which compared to the second precise model, long short term memory predictions are higher between 1% to 14%. Using long short term memory for predicting level of services A and C, support vector machine for predicting level of services B and D and deep neural network for predicting E and F, bring the highest accuracy for each level of service.


Main Subjects