Modeling Urban Venue Dynamics through Spatio-Temporal Metrics and Complex Networks

The ubiquity of GPS-enabled devices, mobile applications, and intelligent transportation systems have enabled opportunities to model the world at an unprecedented scale. Urban environments, in particular, have benefited from new data sources that provide granular representations of activities across space and time. As cities experienced a rise in urbanization, they also faced challenges in managing vehicle levels, congestion, and public transportation systems. Modeling these fast-paced changes through rich data from sources such as taxis, bikes, and trains has enabled prediction models capable of characterizing trends and forecasting future changes. Data-driven studies of urban mobility dynamics have been instrumental in helping deliver more contextual services to cities, support urban policy, and inform business decisions. This dissertation explores how novel algorithmic architectures and techniques reveal and predict business trends and urban development patterns. The research informing this dissertation harnesses principles from network science, modeling cities as connected networks of venues. Building upon a foundation of research in complex network theory, urban computing, and machine learning, we propose algorithms tailored for three computing tasks focused on modeling venue dynamics, characteristics, and trends. First, we predict the demand for newly opened businesses using insights from movement patterns across different regions of the city. Through this analysis we demonstrate how temporally similar areas can be successfully used as inputs to predict the visitation patterns of new venues. Next, we forecast the likelihood of business failure through a supervised learning model. We analyze the value of varying features in predicting business failure and explore their impact across new and established venues and across different cities worldwide. Finally, we present a deep learning architecture which integrates both spatial and topological features to predict the future demand for a venue. These works highlight the power of complex network measures to quantify the structure of a city and inform prediction models. This dissertation leverages vast amounts of data from spatio-temporal networks to model venue dynamics. The research puts forward evidence to support a data-driven study of geographic systems applied to fundamental questions in urban studies, retail development, and social science.

Name dc:type.qualificationname

Doctor of Philosophy (PhD)

Level dc:type.qualificationlevel

Doctoral

Grantor dc:publisher.institution

University of Cambridge

Year dc:date.issued

2020

Author dc:creator

• D'Silva, Krittika

Advisor dc:contributor.advisor

• Mascolo, Cecilia

• Machine learning
• Spatio-temporal modelling

Licence

• https://apollo8-f-pro.lib.cam.ac.uk/bitstreams/5aff2a5c-edd9-4fee-a5fe-b0c913c33738/download
• https://www.rioxx.net/licenses/all-rights-reserved/

Language dc:language

eng

Author Identifier

0000-0002-1411-8523
0000-0001-9614-4380

OAI identifier oai:identifier

oai:www.repository.cam.ac.uk:1810/311355