Citation

Pedreira Junior JU, Assirati L, Pitombo CS. Case Stud. Transp. Policy 2021; 9(4): 1715-1726.

Copyright

(Copyright © 2021, World Conference on Transport Research Society, Publisher Elsevier Publishing)

DOI

10.1016/j.cstp.2021.07.019

PMID

unavailable

Abstract

Important correlations between the morphological characteristics of the urban network and travel behavior can be found in the literature of urban and transport planning. Considering the significant growth in the production of Volunteered Geographic Information, research opportunities in this area are promising. This paper evaluates the usefulness of morphological characteristics of the urban street network of the city of São Carlos (Brazil) by analyzing university students' travel patterns. First, the characteristics of the displacements of these individuals over three consecutive days were obtained by tracking their smartphones. By processing this data, three distinct travel pattern groups were found using the K-Means algorithm, which corresponded to the categories of the dependent variable of the modeling procedure. Then, the coordinates of the residences of these individuals and the street network from the OpenStreetMap geodata were used to generate three types of features associated with: a) the centrality of the place of residence; b) the individual's neighborhood network characteristics; and c) the shortest path between the individuals' dwellings to the campus. Afterwards, these morphological variables and the individual's sociodemographic attributes were used as covariates in a Random Forest Classifier and a Support Vector Machine algorithm to classify the travel patterns previously found in the K-Means algorithm. Two covariate scenarios were evaluated: the first consisting of only sociodemographic attributes and the second comprised by both the sociodemographic attributes and the morphological features. This enabled the evaluation of whether the classification performance was improved with the urban network characteristics. Consistent results were observed, and the average accuracy score increased by 38% with the Random Forest Classifier and by 42% with the Support Vector Machine algorithm. Overall, there is a plausible indication that VGI can be a useful tool in complementing traditional modeling approaches that use travel survey data. From the transport policy perspective, the relationships between these variables and mobility patterns can support the decision-making process regarding urban design and the planning of the transport system.

Language: en

Keywords

Random forest classifier; Support vector machine; Travel pattern analysis; Urban network