uMAD: Classification, Generation and Analysis of User Mobility and Activity Data.

uMAD explores the challenges and solutions related to user mobility and activity (uMA) data. It highlights the importance of uMA data for infrastructure planning, network optimization, and urban development while addressing privacy concerns and data diversity issues.The dissertation introduces a tool designed to classify, generate, evaluate, and analyze uMA datasets. The tool integrates machine learning models like Generative Adversarial Networks (GANs) to create synthetic datasets that mimic real-world mobility traces. The research also presents a taxonomy for classifying mobility datasets based on mobility mode, data sources, and information categories.

The study further examines existing public uMA datasets, their classification, and privacy concerns while proposing synthetic data generation as a privacy-preserving solution. It discusses deep generative models like GANs, VAEs, and HMMs for dataset creation. The experimental results compare various GAN models and analyze their effectiveness in generating realistic traces. The dissertation concludes with future research directions, including improving synthetic data fidelity and expanding uMAD’s applications. It aims to provide an open-source, holistic framework for handling mobility data while addressing privacy and accessibility challenges.