Traffic congestion forecasting using multilayered deep neural network

ABSTRACT

This study proposes a multilayered deep neural network (MLDNN) and a congestion index (CI) based on traffic density factor to forecast traffic congestion directly. Data were collected in Delhi city from a selected location using video cameras during peak hours of weekdays from Monday to Sunday to test the proposed model. Collected data were categorized in a matrix format in the intervals of five-minutes. The input matrix was divided into a number of intervals to train, validate, and test the MLDNN and baseline models, including support vector regression, multi-layer perceptron neural network, gated recurrent unit (GRU) neural network, long short-term memory (LSTM) neural network, convolutional neural network (CNN), CNN-GRU neural network, and CNN-LSTM neural network. Results of the study show that the MLDNN and proposed CI can be applied to predict traffic congestion successfully in heterogeneous traffic.

KEYWORDS:

• Congestion prediction
• Deep learning
• Traffic density
• Congestion index

Acknowledgments

First author is thankful to University Grants Commission (UGC) for providing financial support for this study through the start-up grant research project “Modelling and simulation of vehicular traffic flow problems” through the grant No. F.30-403/2017(BSR). Financial support to the second author form UGC in the form of JRF is also thankfully acknowledged.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Authors’ contributions

Manoj Kumar: Conceptualisation, Methodology, Data collection and Extraction, Original draft preparation, Programming.

Kranti Kumar and Pritikana Das: Review and editing of original draft, supervision, formal analysis, funding acquisition and project administration.

Availability of data and material

Data used in this research were collected by authors under the UGC funded research project ‘Modelling and simulation of vehicular traffic flow problems.’ If there are relevant research needs, the data can be obtained by sending an e-mail to Kranti Kumar ([email protected]).

Code availability

MATHEMATICA and Python were used for programming purpose in this study. Codes can be provided on request for relevant research needs by sending an e-mail to Kranti Kumar ([email protected]). Please indicate the purpose of the research in the e-mail.

Additional information

Funding

This research was supported by University Grants Commission (UGC), New Delhi, India, through the start-up grant research project “Modelling and simulation of vehicular traffic flow problems” via the grant No. F.30-403/2017(BSR).