A deep learning approach to real-time short-term traffic speed prediction with spatial-temporal features

In the realm of Intelligent Transportation Systems (ITS), accurate traffic speed prediction plays an important role in traffic control and management. The study on the prediction of traffic speed has attracted considerable attention from many researchers in this field in the past three decades. In recent years, deep learning-based methods have demonstrated their competitiveness to the time series analysis which is an essential part of traffic prediction. These methods can efficiently capture the complex spatial dependency on road networks and non-linear traffic conditions. We have adopted the convolutional neural network-based deep learning approach to traffic speed prediction in our setting, based on its capability of handling multi-dimensional data efficiently. In practice,the traffic data may not be recorded with a regular interval, due to many factors, like power failure, transmission errors,etc.,that could have an impact on the data collection. Given that some part of our dataset contains a large amount of missing values, we study the effectiveness of a multi-view approach to imputing the missing values so that various prediction models can apply. Experimental results showed that the performance of the traffic speed prediction model improved significantly after imputing the missing values with a multi-view approach, where the missing ratio is up to 50%.

Name thesis:degree_name

M.Sc.

Level thesis:degree_level

Masters

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Windsor

Year dc:date.issued

2019

Author dc:creator

• Gutha, Sindhuja

Advisors dc:contributor.advisor

• Jessica Chen
• Mehdi Kargar

Contributors dc:contributor

• mita@uwindsor.ca

Licence

• http://creativecommons.org/licenses/by-nc-nd/4.0/

Language dc:language.iso

en_CA

Handle dc:identifier.uri

https://hdl.handle.net/20.500.14776/8028

OAI identifier oai:identifier

oai:uwindsor.scholaris.ca:20.500.14776/8028