University of Twente Student Theses


Traffic prediction based on probabilistic graphical method

Jayashankaramma Shankar Raj, Mohan Raju (2020) Traffic prediction based on probabilistic graphical method.

Link to full-text:
(only accessible for UT students and staff)
Abstract:Road traffic congestion is considered as a ubiquitous, a phenomenon which occurs during peak hours in densely covered areas like in Amsterdam and Rotterdam. This includes the complexity of planning infrastructure for transportation. One of the critical issues involved is to identify design elements (Traffic lanes, road type and driving lane). On the other hand, the road network in an urban area is a distinctive network. The study of such systems includes the evaluation and optimization of the traffic state. Needless to say, more traffic on the outer roads has a more significant impact on the traffic condition within the city: A considerable deal of start and end of traffic journey on the urban road network. Here the question arises is how traffic was developed over time within the given network? This research proposes a novel framework which is a threefold approach to address the traffic congestion on the Dutch highways. The entire framework is developed based on a Bayesian network (BN) approach. The network structure was modelled based on the probabilistic dependency, which helps in identifying causes for congestion. Firstly, the BN was established over the set of random variables which is called as ‘contributing variables’ such as Time of the day, Speed and vehicle characteristics. The discrete variables are further evaluated individually to identify the major causes of traffic congestion. Furthermore, the association between contributing variables and congestion occurrence was quantified by measuring the odds ratio. Secondly, using the same network model, the traffic flow predictions are performed by using continuous variables. For the continuous variables, the data was downloaded for 30 & 60-minute intervals, respectively. The accuracy of the result of the mentioned 2-time intervals was measured based on RMSE, MAE and MAPE. Thirdly, the characteristics of Amsterdam and Rotterdam road network was studied based on the qualitative and quantitative aspect. A qualitative aspect includes the visual inspection that describes the texture and gradient of the road network. Quantitative aspect involves the descriptive statistics that describe node density, street length, number of nodes and average street length. The entire study made use of two different data sources. • For prediction and congestion diagnosis- historical dataset extracted from NDW, which is an open-access database which attempts to collect, process, store and distribute all the traffic-related data. • For analysing road network characteristics – OSM database was considered. The results from the BN model states that Firstly - the contributing variables such as Travel Time, Time of the day and speed as a major impact on traffic congestion occurrence. Secondly – 30 minutes predictions are better, as the RMSE value was recorded lower compare to 60 minutes interval. Thirdly, based on the road network statistics - Amsterdam as many nodes, edges and networks are in semi-structured type and Rotterdam has fewer number of edges and nodes with square typed structure.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page