Predicting CO₂ emissions of private transport in Helsinki region using transfer learning

Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

School of Engineering | Master's thesis

Authors

Date

2025-07-31

Department

Major/Subject

Geoinformatics

Mcode

Degree programme

Master's Programme in Geoinformatics

Language

en

Pages

52

Series

Abstract

The thesis explores the application of transfer learning to forecast the CO2 emissions of the private transport in Helsinki in laboratory conditions, using models trained on detailed trip data as well as built environment data provided by Espoo. To develop a robust dataset, the research incorporates Locomizer origin-destination mobility records based on H3 hexagon grids to achieve a fine spatial resolution, Statistics Finland grid data, and transport infrastructure parameters. The Espoo data underwent training of three machine learning models, Random Forest, Hist Gradient Boosting, and Linear Regression, and the models were applied to Helsinki data, with Random Forest being the most accurate at predicting (R > 0.98). It is seen in the analysis that the total car distance, shortest network path, and the frequency of trips are the key predictors of emissions. Findings demonstrate that the practical applicability of transfer learning on urban emission modelling in regions with limited data availability, which would enhance high resolution modelling scalability and policymaking, as well as planning. This methodology and results will promote the creation of data- driven, transferable tools in climate action in urban areas.

Description

Supervisor

Tenkanen, Henrikki

Thesis advisor

Dey, Subhrasankha

Keywords

CO₂ emissions, private transport, transfer learning, emission prediction, H3 hexagons, random forest

Other note

Citation

Permanent link to this item

https://urn.fi/URN:NBN:fi:aalto-202508196345

Collections

[dipl] Insinööritieteiden korkeakoulu / ENG