Ph.D. Defense: Xi Yang

Title: Spatio-Temporal Human Mobility Analytics and Prediction

Ph.D. Candidate: Xi Yang

Major Advisor: Dr. Suining He
Advisory Committee:  Dr. Bing Wang, Dr. Dongjin Song
Review Committee:  Dr. Derek Aguiar, Dr. Minmei Wang
Date/Time: Thursday, May 11th, 2023, 10:00 am

Location: WebEx

Meeting link:

https://uconn-cmr.webex.com/uconn-cmr/j.php?MTID=m56f7a0275587570bf2a590092118c61c

Meeting number:

2624 001 4436

Password:

ujSmSMCF288

Abstract 

In this thesis defense, we aim at investigating ubiquitous human mobility analytics and accurate prediction techniques of human mobility patterns (e.g., crowd flows and transitions across locations, and bike sharing usage). Accurate spatio-temporal human mobility analytics and prediction approaches can enable various ubiquitous and urban computing applications, such as event surveillance, urban planning and operation, epidemic and social behavior analysis, and location-based recommendation. Toward the spatio-temporal human mobility analysis, this thesis defense will focus on the following three challenges: (a) the harvested crowd mobility data may have severe sparsity and skewness issue, i.e. the majority of the crowd mobility data (in/out flows and transitions) are recorded at a few locations and in a few time periods; (b) the human mobility sensing infrastructure (e.g., the wireless infrastructure) may be altered due to the needs of maintenance and re-installation, leading to different mobility patterns; (c) due to the users’ complex daily routines and preferences, the mobility patterns at fine granularity (such as pick-ups/drop-offs for each bike sharing station) vary spatially and temporarily across different locations and time periods and can be complicated to forecast.

In this thesis, we conduct human mobility pattern case studies upon the campus Wi-Fi based crowd sensing infrastructure and the city-wide bike-sharing system, for the following three interconnected research thrusts. Based on the campus Wi-Fi association data, we have studied the data sparsity and skewness and jointly predict crowd flows (e.g., in- and out-flows at different positions) and transitions (e.g., movement across the locations) by an attention-based graph embedding design. To overcome the challenge of altered human mobility sensing infrastructure, we have further designed an adaptive learning and modeling approach to predict the crowd flow distributions at locations with altered mobility sensing infrastructure (e.g., installation of new Wi-Fi access points). To enable the accurate and fine-grained human mobility prediction, we have designed a graph adjacency attention neural network to predict station-level bike traffic (bike pick-ups and drop-offs) for entire metropolitan bike sharing systems.