Files in this item

FilesDescriptionFormat

application/pdf

application/pdfXU-DISSERTATION-2019.pdf (2MB)Restricted Access
(no description provided)PDF

Description

Title:Advanced space-time integration for knowledge discovery in human mobility studies
Author(s):Xu, Li
Director of Research:Kwan, Mei-Po
Doctoral Committee Chair(s):Kwan, Mei-Po
Doctoral Committee Member(s):McLafferty, Sara; Li, Bo; Cidell, Julie
Department / Program:Geography & Geographic InfoSci
Discipline:Geography
Degree Granting Institution:University of Illinois at Urbana-Champaign
Degree:Ph.D.
Genre:Dissertation
Subject(s):Space-time modeling
Mobility
Space-time kernel
Sequential pattern mining
Bayesian networks
Demand modeling
Deep learning
LSTM
Abstract:In the past decade or so, advances in positioning technologies and the prevalence of smart personal devices for capturing individual movement have given rise to a wide range of studies, including transportation, public health, tourism, and social network services. With the fast-growing volume of and interest in spatio-temporal mobility data, there is also an increasing need for new methods of analyzing this kind of data. Particularly, considerable effort has been made to characterize human activity-travel patterns from the spatio-temporal mobility data. However, knowledge discovery from the large-scale human mobility data remains a challenging task due to its complex spatio-temporal variations. Most traditional statistical and machine learning techniques are powerful for prediction tasks but are not tailored to characterize the dynamical spatio-temporal correlations in the human mobility data. This dissertation offers several methodological and practical contributions to the field through developing a set of novel methods and validation with real-world use cases. First, a locally adaptive space-time kernel approach is proposed to model the non-emergency municipal services demand in Chicago. Second, this work develops novel sequential similarity measures for analyzing human activity-travel patterns and Bayesian networks models with specially designed topology to predict the forthcoming activity at the individual level. Last, a deep convolutional LSTM networks model is proposed to capture the spatial and temporal dependencies in an integrated way to predict real-time taxi demand. All proposed models are proven to be more effective or robust in various real-world experiments as compared to traditional statistical or machine learning algorithms but are not limited to these use cases. The methods contribute to any point demand modeling, regional demand modeling, and sequential trajectory learning problems for spatio-temporal data.
Issue Date:2019-12-04
Type:Text
URI:http://hdl.handle.net/2142/106442
Rights Information:Copyright 2019 Li Xu
Date Available in IDEALS:2020-03-02
Date Deposited:2019-12


This item appears in the following Collection(s)

Item Statistics