Generation of Realistic Road Speed Data Using Generative Adversarial Networks

The analysis and testing of smart ecosystems, such as smart cities, require a huge quantity
of data from the field. However, the lack of open data is a common problem that hinders
the ability to perform accurate simulations and predictions. Therefore, synthetic data can be
used to test smart ecosystems in the absence of real data. Generating synthetic datasets that
faithfully reproduce real datasets is an open problem. Generative Adversarial Networks (GANs)
are promising deep learning models that provide a mechanism for creating realistic randomised
datasets. Contextualised in a ride-sharing ecosystem, the thesis aims at developing and training
a GAN network for generating synthetic traffic data and evaluating its fitness, given a dataset
of real traffic data downloaded from the official Uber Movement platform.