Exportar Publicação

Bhosale, V., San Payo, M., Cipriano, G. & Andrade, A. R. (2025). Key factors affecting transportation choices in school commuting in Lisbon: A machine learning approach. Case Studies on Transport Policy. 21

V. Bhosale et al.,  "Key factors affecting transportation choices in school commuting in Lisbon: A machine learning approach", in Case Studies on Transport Policy, vol. 21, 2025

@article{bhosale2025_1764928307518,
	author = "Bhosale, V. and San Payo, M. and Cipriano, G. and Andrade, A. R.",
	title = "Key factors affecting transportation choices in school commuting in Lisbon: A machine learning approach",
	journal = "Case Studies on Transport Policy",
	year = "2025",
	volume = "21",
	number = "",
	doi = "10.1016/j.cstp.2025.101557",
	url = "https://www.sciencedirect.com/journal/case-studies-on-transport-policy"
}

TY  - JOUR
TI  - Key factors affecting transportation choices in school commuting in Lisbon: A machine learning approach
T2  - Case Studies on Transport Policy
VL  - 21
AU  - Bhosale, V.
AU  - San Payo, M.
AU  - Cipriano, G.
AU  - Andrade, A. R.
PY  - 2025
SN  - 2213-624X
DO  - 10.1016/j.cstp.2025.101557
UR  - https://www.sciencedirect.com/journal/case-studies-on-transport-policy
AB  - Understanding the behaviour of students in choosing a transportation mode to school is crucial to promote Active Commuting to School (ACS) and the adoption of healthier lifestyles. Therefore, analysing all types of transportation modes with multiple factors/features is essential, though it can be a challenge in statistical modelling. The main objective of the present study was to determine the factors that contribute to the choice of a particular mode in school transportation, by using Machine Learning (ML) algorithms: Extreme Gradient Boosting (XGB), Random Forest (RF), Decision Tree (DT) and Multinomial Logistic Regression (MNL). Data from the ‘Hands Up’ Survey in Lisbon, Portugal, between 2018 and 2021, with 10 different modes of transportation were analysed. A range of factors including safety around school, socioeconomic status of schools’ parishes, school regime, school grades and the proximity of schools to the different public transportation modes were considered. The algorithms have been compared in terms of accuracy scores. The XGB algorithm shows the best performance (64 % accuracy and 0.33 Macro F1) for multi-class classification, while RT, DT and MNL provide accuracy of 40 %, 37 % and 47 % respectively. Weighted Average Feature Importance (WAFI) have been determined for all variables. For the best-performing algorithm, the XGB, the combination factor of school regime and school grade is the most relevant factor, contributing to around 21.2 % for multi-class classification. WAFI scores for each variable suggest that the proximity of schools to various public transports is an important factor contributing more than 50 % for the predominance of private car in school transportation.
ER  -