Show simple item record

dc.contributor.authorRahi, Arsalan Ahmad
dc.date.accessioned2020-04-16T08:49:44Z
dc.date.available2020-04-16T08:49:44Z
dc.date.issued2019-10-23
dc.identifier.urihttp://hdl.handle.net/2299/22590
dc.description.abstractIntelligent Transport Systems (ITS) as a field has emerged quite rapidly in the recent years. A competitive solution coupled with big data gathered for ITS applications needs the latest AI to drive the ITS for the smart and effective public transport planning and management. Although there is a strong need for ITS applications like Advanced Route Planning (ARP) and Traffic Control Systems (TCS) to take the charge and require the minimum of possible human interventions. This thesis develops the models that can predict the traffic link flows on a junction level such as road traffic flows for a freeway or highway road for all traffic conditions. The research first reviews the state-of-the-art time series data prediction techniques with a deep focus in the field of transport Engineering along with the existing statistical and machine leaning methods and their applications for the freeway traffic flow prediction. This review setup a firm work focussed on the view point to look for the superiority in term of prediction performance of individual statistical or machine learning models over another. A detailed theoretical attention has been given, to learn the structure and working of individual chosen prediction models, in relation to the traffic flow data. In modelling the traffic flows from the real-world Highway England (HE) gathered dataset, a traffic flow objective function for highway road prediction models is proposed in a 3-stage framework including the topological breakdown of traffic network into virtual patches, further into nodes and to the basic links flow profiles behaviour estimations. The proposed objective function is tested with ten different prediction models including the statistical, shallow and deep learning constructed hybrid models for bi-directional links flow prediction methods. The effectiveness of the proposed objective function greatly enhances the accuracy of traffic flow prediction, regardless of the machine learning model used. The proposed prediction objective function base framework gives a new approach to model the traffic network to better understand the unknown traffic flow waves and the resulting congestions caused on a junction level. In addition, the results of applied Machine Learning models indicate that RNN variant LSTMs based models in conjunction with neural networks and Deep CNNs, when applied through the proposed objective function, outperforms other chosen machine learning methods for link flow predictions. The experimentation based practical findings reveal that to arrive at an efficient, robust, offline and accurate prediction model apart from feeding the ML mode with the correct representation of the network data, attention should be paid to the deep learning model structure, data pre-processing (i.e. normalisation) and the error matrices used for data behavioural learning. The proposed framework, in future can be utilised to address one of the main aims of the smart transport systems i.e. to reduce the error rates in network wide congestion predictions and the inflicted general traffic travel time delays in real-time.en_US
dc.language.isoenen_US
dc.rightsinfo:eu-repo/semantics/openAccessen_US
dc.rightsAttribution 3.0 United States*
dc.rights.urihttp://creativecommons.org/licenses/by/3.0/us/*
dc.subjectAIen_US
dc.subjectMachine Learningen_US
dc.subjectOptimisationen_US
dc.subjectTraffic Predictionsen_US
dc.subjectRoad Predictionsen_US
dc.subjectTraffic Flow Forecastingen_US
dc.subjectGRUen_US
dc.subjectRNNen_US
dc.subjectLSTMen_US
dc.subjectNeural Networken_US
dc.subjectSVMen_US
dc.subjectRandom Foresten_US
dc.subjectDBNen_US
dc.subjectBack Propagationen_US
dc.subjectARIMAen_US
dc.subjectRegression Predictionen_US
dc.subjectGrid Searchen_US
dc.subjectCorrelation Analysisen_US
dc.subjectJunction Levelen_US
dc.subjectNode Predictionen_US
dc.subjectActivation Functionen_US
dc.titleMachine Learning Approaches for Traffic Flow Forecastingen_US
dc.typeinfo:eu-repo/semantics/masterThesisen_US
dc.identifier.doidoi:10.18745/th.22590*
dc.identifier.doi10.18745/th.22590
dc.type.qualificationlevelMastersen_US
dc.type.qualificationnameMPhilen_US
dcterms.dateAccepted2019-10-23
rioxxterms.funderDefault funderen_US
rioxxterms.identifier.projectDefault projecten_US
rioxxterms.versionNAen_US
rioxxterms.licenseref.urihttps://creativecommons.org/licenses/by/4.0/en_US
rioxxterms.licenseref.startdate2020-04-16
herts.preservation.rarelyaccessedtrue
rioxxterms.funder.projectba3b3abd-b137-4d1d-949a-23012ce7d7b9en_US


Files in this item

Thumbnail
Thumbnail

This item appears in the following Collection(s)

Show simple item record

info:eu-repo/semantics/openAccess
Except where otherwise noted, this item's license is described as info:eu-repo/semantics/openAccess