Título:	Using machine learning to predict activity chains and mode choice on transportation models
Autor(es):	Miranda, Daniele Firme
E-mail do autor:	danielefirme@gmail.com danielefirme@unb.br
Orientador(es):	Gonzales Taco, Pastor Willy
Assunto:	Modelagem de transportes Aprendizagem de máquina Mobilidade urbana
Data de publicação:	22-Mar-2021
Data de defesa:	26-Out-2020
Referência:	MIRANDA, Daniele Firme. Using machine learning to predict activity chains and mode choice on transportation models. 2020. xii, 96 f., il. Dissertação (Mestrado em Transportes)—Universidade de Brasília, Brasília, 2020.
Abstract:	When travel is considered a demand derived from people’s need to perform activities, it becomes clear that a better understanding of how people organize their activities during a day must provide a more solid basis for travel demand modeling. By replicating disaggregate travel decisions (at the individual level), activity-based models may produce better travel demand predictions, compared to the previous generations of modeling approaches (tripbased approaches, for instance). A paper published in 2019 stands out among the most recent activity-based modeling research as the authors propose a comprehensive framework for generating full and detailed activity schedules for given agents depending on their sociodemographic features, called Data-Driven Activity Scheduler (DDAS). The aim of this research was to develop a commented replication of the methodological approach of two modules of the DDAS: the Activity Type Model (ATM) and the Mode Choice Model (MCM). Specific objectives included replicating these two modules of the DDAS framework using data from the Federal District Urban Mobility Survey, which is significantly larger than the dataset used in the original DDAS study. Moreover, it was intended to investigate possible improvements to be made on the DDAS framework, including its validation procedure. The obtained results from the replication of the DDAS framework indicated that there was improvement to be made on the manner how models were being trained, in order to better deal with class imbalance. Therefore, a second implementation was made by using the SMOTE technique (Synthetic Minority Oversampling Technique) for training the ATM and MCM modules. Although activity chains seemed more realistic in this second set of results, the overall validation score for the ATM module was low. Therefore, a third model was developed by training the models as Random Forest classifiers instead of isolated Decision Tree classifiers as it was defined in the original DDAS framework. Significant improvement was observed in the results of this third model, both in training and test, for both ATM and MCM modules. Furthermore, another contribution of this study is the public availability of all scripts that were developed during its conduction.
Unidade Acadêmica:	Faculdade de Tecnologia (FT) Departamento de Engenharia Civil e Ambiental (FT ENC)
Informações adicionais:	Dissertação (mestrado)—Universidade de Brasília, Faculdade de Tecnologia, Departamento de Engenharia Civil e Ambiental, 2020.
Programa de pós-graduação:	Programa de Pós-Graduação em Transportes
Licença:	A concessão da licença deste item refere-se ao termo de autorização impresso assinado pelo autor com as seguintes condições: Na qualidade de titular dos direitos de autor da publicação, autorizo a Universidade de Brasília e o IBICT a disponibilizar por meio dos sites www.bce.unb.br, www.ibict.br, http://hercules.vtls.com/cgi-bin/ndltd/chameleon?lng=pt&skin=ndltd sem ressarcimento dos direitos autorais, de acordo com a Lei nº 9610/98, o texto integral da obra disponibilizada, conforme permissões assinaladas, para fins de leitura, impressão e/ou download, a título de divulgação da produção científica brasileira, a partir desta data.
Aparece nas coleções:	Teses, dissertações e produtos pós-doutorado

Os itens no repositório estão protegidos por copyright, com todos os direitos reservados, salvo quando é indicado o contrário.