Abstract

Traffic congestion is a widely occurring phenomenon caused by increased use of vehicles on roads resulting in slower speeds, longer delays, and increased vehicular queueing in traffic. Every year, over a thousand hours are spent in traffic congestion leading to great cost and time losses. In this thesis, we propose a multimodal data fusion framework for predicting traffic congestion on urban motorway networks. It comprises of three main approaches. The first approach predicts traffic congestion on urban motorway networks using data mining techniques. Two categories of models are considered namely neural networks, and random forest classifiers. The neural network models include the back propagation neural network and deep belief network. The second approach predicts traffic congestion using social media data. Twitter traffic delay tweets are analyzed using sentiment analysis and cluster classification for traffic flow prediction.

Lastly, we propose a data fusion framework as the third approach. It comprises of two main techniques. The homogeneous data fusion technique fuses data of same types (quantitative or numeric) estimated using machine learning algorithms. The heterogeneous data fusion technique fuses the quantitative data obtained from the homogeneous data fusion model and the qualitative or categorical data (i.e. traffic tweet information) from twitter data source using Mamdani fuzzy rule inferencing systems.

The proposed work has strong practical applicability and can be used by traffic planners and decision makers in traffic congestion monitoring, prediction and route generation under disruption.