Available at: https://digitalcommons.calpoly.edu/theses/3352
Date of Award
6-2026
Degree Name
MS in Computer Science
Department/Program
Computer Science
College
College of Engineering
Advisor
Theresa Migler
Advisor Department
Computer Science
Advisor College
College of Engineering
Abstract
This thesis evaluates the relationships between various graph theory metrics and taxi traffic volume for the cities of San Francisco, California and Porto, Portugal. We also evaluate a modified betweenness centrality metric which incorporates the count of distinct origin-destination pairs from the taxi data as the weight function. This thesis extends a paper by Pengyao Ye, Bo Wu, and Wenbo Fan by reducing circularity through a temporal train-test split and by comparing both line-graph and primal-graph formulations of betweenness centrality.
We found that past traffic volume is almost perfectly correlated with future traffic volume and that the modified betweenness centrality substantially outperforms purely structural metrics. We also found that weighted primal-graph formulations of betweenness centrality consistently outperform the line-graph formulation used in the original paper and, in the San Francisco case study, outperform raw origin-destination counts. These results suggest traffic patterns are highly temporally consistent, that degree centrality, closeness centrality, and pure betweenness centrality poorly correlate with traffic volume, and that preserving street-length information in the primal graph may improve the predictive power of modified betweenness centrality.