Transfer Learning Solutions for Intelligent Transportation

The long-established infrastructure for cities is being strained to the point of breaking. Cities are no longer state of the art and with a growing population they have to manage widespread pollution, traffic congestion and the rising costs of maintaining and powering inefficient energy infrastructure. The transition into Smart Cities shows promise in rectifying these issues because it's a technological approach to road traffic management which replaces traditional traffic monitoring and control with a more eco-friendly and cost-effective approach. This research project aims to measurably (and comprehensively) improve transportation related decision making with regards to traffic monitoring and control.
The project will investigate and analyse current modelling and prediction approaches including time series, deep learning, and evolutionary computation as well as the potential of transfer learning in large-scale urban road network modelling. The investigation will involve the implementation of these approaches which when completed will be used to build robust, accurate, and reusable computational models of traffic flow through major cities. Throughout the research project, the UK's Department for Transport's traffic data will be extracted to train prediction models. The data extraction process will require a data pre-processing step where data are divided into training sets to train the predictive models as well as testing and validation sets which will be used in the evaluation process.
The evaluation process will consist of model validation, experimentation, and testing initially using the Birmingham traffic data, after it has been appropriately divided, in order to assess the validity and accuracy of prediction models. Furthermore, depending on the predictive modelling approach, cross-validation or nested cross-validation can be applied during the evaluation. The investigation and evaluation of current modelling and prediction techniques, alongside their application, are the groundwork for an original contribution to science which will be a set of new, accurate, robust, scalable, and economical urban traffic prediction algorithms.
The research project's proposed stakeholders are ASTUTE, Birmingham City Council, and the UK's Department for Transport. The principal project beneficiary however is the Birmingham City Council as the project aims to have the output from the developed prediction algorithms be integrated effectively with the city council policies to measurably improve people's lives. The involvement of city authorities is a unique strength for this project as the majority of relevant research proposed has not been effectively integrated and utilised by municipalities. This project intends to go further than just contributing to science by consolidating a collaborative relationship with stakeholders and measurably improving people's lives. The collaborative relationship will be one that follows an agile based approach where the project's stakeholders will help outline the project's product backlog, evaluate the progress made and give feedback throughout development.