Implementing deep learning techniques for network-scale traffic forecasting

In the past few years, Deep learning has re-emerged as a powerful tool to solve complex problems and create prediction models that can outperform a lot of the existing state-of-the-art methods. This is primarily due to two main reasons; the rise of big data, where huge amounts of information has become readily available to the public, as well as the recent technological advancements in computer processing powers which has enabled researchers to take advantage of these large volumes of data. One of the major fields which requires dealing with and understanding extensive amounts of data is transportation. In the United States alone, 220 billion vehicle trips have taken place in 2017 [1]. This creates the need for researchers who can work with such huge data to build models and infer beneficial knowledge which can contribute to improving transportation networks and the overall travel experience. In this thesis, we study the use of several machine learning and deep learning techniques to predict travel times on a road network. The two main methods proposed to tackle the problem are Convolutional Neural Networks and Long-Short Term Memory Networks. The location of interest of this thesis is the city of New York. The New York City Taxi and Limousine Commission provides the origin and destination pairs, along with the travel times and other formation, for each taxi trip between the years of 2010 and 2013. A more refined representation of the data was obtained from B. Donovan and D. Work [2], where travel time estimates for each hour of the day is provided along every road in the city.U of I OnlyAuthor requested U of Illinois access only (OA after 2yrs) in Vireo ETD syste