The provision of accurate and timely information to traffic analysts and road users are critical components for the successful implementation of intelligent transportation systems (ITSs) around the world. This is typically achieved via the application of predictive analytics on historical data to make forecasts about traffic parameters. However, given the broad spectrum of data sources, data collection methods and traffic predictive models at the disposal of traffic data scientists, making accurate predictions becomes challenging for some reasons. Firstly, the complexity of the traffic domain makes traffic predictive analytics (TPA) problem description complicated. Secondly, the plethora of available predictive models makes the choice of which model to be applied in each TPA scenario difficult. Thirdly, there is not yet a predictive method that works well over time and in all scenarios (Joyce and Herrmann, 2018; Zhang et al., 2019). Due to these limitations, there is a need for the provision of guidance to traffic data scientists performing data-driven traffic prediction. Traffic Predictive Analytics Guidance Framework (TAG-F) is a guidance framework that aims at bridging this gap. The framework delineates data-driven traffic prediction as a set of three dimensions, thereby providing a structured collection of analytical decision points that can serve as a roadmap to enable the traffic data scientist traverse from the traffic problem space to the analytical solution space, culminating in an action/outcome, usually prediction. TAG-F ÃÂ¢ÃÂÃÂ via the tool ÃÂ¢ÃÂÃÂ can also be used to provide decision support for traffic data scientists by providing guidance in the choice of predictive analytical method (PAM), given the data context specifications. The framework and tool were evaluated using real-world traffic prediction scenarios in an urban arterial in Greater Manchester, United Kingdom. The contributions made through the study include a novel end-to-end guidance mechanism for TPA using a framework that fosters a structured definition of the TPA solution development process. In addition, the identification of a set of key dimensions and parameters that influence TPA. A prototype support tool is also presented, which complements the framework by providing semi-automated guidance by suggesting alternative predictive models to given TPA scenarios. The framework and tool can foster productivity in the TPA process by encouraging adaptability, reuse, and shared domain knowledge about TPA. Results from empirical analysis support the value of the proposed framework and support tool towards the provision of guidance to traffic data scientists in TPA, however with some limitations. Finally, in this thesis, suggestions about furthering the study, addressing the identified limitations, and refining the framework and tool are articulated.
|Date of Award||31 Dec 2014|
- The University of Manchester
|Supervisor||Pedro Sampaio (Supervisor)|