2023
Feng, Tao
Machine learning approaches in modeling behavior and for prediction in urban research Presentation
Wuhan University of Technology, 29.12.2023.
Links | BibTeX | Tags: Machine learning
@misc{nokey,
title = {Machine learning approaches in modeling behavior and for prediction in urban research},
author = {Tao Feng},
url = {https://home.hiroshima-u.ac.jp/taofeng},
year = {2023},
date = {2023-12-29},
howpublished = {Wuhan University of Technology},
keywords = {Machine learning},
pubstate = {published},
tppubtype = {presentation}
}
Wu, Jishi; Jia, Peng; Feng, Tao; Li, Haijiang; Kuang, Haibo
Spatiotemporal analysis of built environment restrained traffic carbon emissions and policy implications Journal Article
In: Transportation Research Part D: Transport and Environment, vol. 121, pp. 103839, 2023, ISSN: 1361-9209.
Abstract | Links | BibTeX | Tags: Built environment, GPBoost, Machine learning, Nonlinear effects, SHapley Additive ExPlanation, Traffic carbon emissions
@article{WU2023103839,
title = {Spatiotemporal analysis of built environment restrained traffic carbon emissions and policy implications},
author = {Jishi Wu and Peng Jia and Tao Feng and Haijiang Li and Haibo Kuang},
url = {https://www.sciencedirect.com/science/article/pii/S1361920923002365},
doi = {https://doi.org/10.1016/j.trd.2023.103839},
issn = {1361-9209},
year = {2023},
date = {2023-01-01},
journal = {Transportation Research Part D: Transport and Environment},
volume = {121},
pages = {103839},
abstract = {Urban environmental policies need to be rectified considering the spatioemporal variations of traffic emissions. However, knowledge to support such a decision-making process is insufficient. This study analyzes the spatiotemporal distributions of traffic emissions in the built environment and their potential nonlinear associations. Considering the recent innovations in machine learning, a tree-boosting algorithm combined with Gaussian process and random effects models (GPBoost) is applied using the big GPS taxi data from Dalian, China. The nonlinear relationships between built environment variables and traffic carbon (CO2) emissions are interpreted using the SHapley Additive ExPlanation (SHAP). It is found that the proposed GPBoost model that considers spatial heterogeneity enhances the overall predictive power compared to traditional machine learning models. Most of the built environment variables have a nonlinear relationship with traffic carbon emissions and the threshold effects vary over time, indicating the necessity of dynamic urban management.},
keywords = {Built environment, GPBoost, Machine learning, Nonlinear effects, SHapley Additive ExPlanation, Traffic carbon emissions},
pubstate = {published},
tppubtype = {article}
}
Urban environmental policies need to be rectified considering the spatioemporal variations of traffic emissions. However, knowledge to support such a decision-making process is insufficient. This study analyzes the spatiotemporal distributions of traffic emissions in the built environment and their potential nonlinear associations. Considering the recent innovations in machine learning, a tree-boosting algorithm combined with Gaussian process and random effects models (GPBoost) is applied using the big GPS taxi data from Dalian, China. The nonlinear relationships between built environment variables and traffic carbon (CO2) emissions are interpreted using the SHapley Additive ExPlanation (SHAP). It is found that the proposed GPBoost model that considers spatial heterogeneity enhances the overall predictive power compared to traditional machine learning models. Most of the built environment variables have a nonlinear relationship with traffic carbon emissions and the threshold effects vary over time, indicating the necessity of dynamic urban management.
2022
Yu, Liang; Feng, Tao; Li, Tie; Cheng, Lei
Demand Prediction and Optimal Allocation of Shared Bikes Around Urban Rail Transit Stations Journal Article
In: Urban Transit Rail, 2022.
Abstract | Links | BibTeX | Tags: Big data, Bike sharing, Machine learning, Transit
@article{nokey,
title = {Demand Prediction and Optimal Allocation of Shared Bikes Around Urban Rail Transit Stations},
author = {Liang Yu and Tao Feng and Tie Li and Lei Cheng},
url = {https://link.springer.com/content/pdf/10.1007/s40864-022-00183-w.pdf?pdf=button},
doi = {https://doi.org/10.1007/s40864-022-00183-w},
year = {2022},
date = {2022-12-13},
journal = {Urban Transit Rail},
abstract = {The imbalance between the supply and demand of shared bikes is prominent in many urban rail transit stations, which urgently requires an efficient vehicle deployment strategy. In this paper, we propose an integrated model to optimize the deployment of shared bikes around urban rail transit stations, incorporating a seasonal autoregressive integrated moving average with long short-term memory (SARIMA-LSTM) hybrid model that is used to predict the heterogeneous demand for shared bikes in space and time. The shared bike deployment strategy was formulated based on the actual deployment process and under the principle of cost minimization involving labor and transportation. The model is applied using the big data of shared bikes in Xicheng District, Beijing. Results show that the SARIMA-LSTM hybrid model has great advantages in predicting the demand for shared bikes. The proposed allocation strategy provides a new way to solve the imbalance challenge between the supply and demand of shared bikes and contributes to the development of a sustainable transportation system.},
keywords = {Big data, Bike sharing, Machine learning, Transit},
pubstate = {published},
tppubtype = {article}
}
The imbalance between the supply and demand of shared bikes is prominent in many urban rail transit stations, which urgently requires an efficient vehicle deployment strategy. In this paper, we propose an integrated model to optimize the deployment of shared bikes around urban rail transit stations, incorporating a seasonal autoregressive integrated moving average with long short-term memory (SARIMA-LSTM) hybrid model that is used to predict the heterogeneous demand for shared bikes in space and time. The shared bike deployment strategy was formulated based on the actual deployment process and under the principle of cost minimization involving labor and transportation. The model is applied using the big data of shared bikes in Xicheng District, Beijing. Results show that the SARIMA-LSTM hybrid model has great advantages in predicting the demand for shared bikes. The proposed allocation strategy provides a new way to solve the imbalance challenge between the supply and demand of shared bikes and contributes to the development of a sustainable transportation system.
Li, Xiaodong; Feng, Tao; Rasouli, Soora
Exploring random taste heterogeneity in choice modelling using mixture density network Conference
7th International Choice Modelling Conference (ICMC), May 23-25, 2022. Reykjavik, Iceland, 2022.
Abstract | Links | BibTeX | Tags: Choice models, Heterogeneity, Machine learning
@conference{8ee855dfda764e948993b1c62df890a4,
title = {Exploring random taste heterogeneity in choice modelling using mixture density network},
author = {Xiaodong Li and Tao Feng and Soora Rasouli},
url = {http://www.icmconference.org.uk/2022-icmc-reykjavik.html},
year = {2022},
date = {2022-01-31},
urldate = {2022-01-31},
address = {Reykjavik, Iceland},
organization = {7th International Choice Modelling Conference (ICMC), May 23-25, 2022.},
abstract = {Capturing heterogeneity in subjects’ decision making process, as accurate as possible, plays an essential role in choice modeling research. In this paper, we investigate the random taste heterogeneity in travel behavior modeling which is an integral part of decision-making process. In contrast to previous works, we use the Mixture Density Network (MDN) which is built from Neural Network and mixture Gaussian model to identify the latent heterogeneity. We assume that the taste variation of individuals follows a series of distribution with certain mean and standard deviation which are dependent on individual social demographic characteristics. We integrated this machine learning method into the discrete choice model and jointly estimated the parameters. Using the stated preference data of Swissmetro, we applied our proposed model and discovered random taste variations which are highly interpretable. We also compared the model with traditional mixed logit model and found the superiority of the proposed model.},
keywords = {Choice models, Heterogeneity, Machine learning},
pubstate = {published},
tppubtype = {conference}
}
Capturing heterogeneity in subjects’ decision making process, as accurate as possible, plays an essential role in choice modeling research. In this paper, we investigate the random taste heterogeneity in travel behavior modeling which is an integral part of decision-making process. In contrast to previous works, we use the Mixture Density Network (MDN) which is built from Neural Network and mixture Gaussian model to identify the latent heterogeneity. We assume that the taste variation of individuals follows a series of distribution with certain mean and standard deviation which are dependent on individual social demographic characteristics. We integrated this machine learning method into the discrete choice model and jointly estimated the parameters. Using the stated preference data of Swissmetro, we applied our proposed model and discovered random taste variations which are highly interpretable. We also compared the model with traditional mixed logit model and found the superiority of the proposed model.
2011
Feng, T; Arentze, T A; Timmermans, H J P
Assessing the relative importance of input variables for route choice modeling: a neural network approach Journal Article
In: Journal of the Eastern Asia Society for Transportation Studies, vol. 9, pp. 341-353, 2011, ISSN: 1881-1124.
Abstract | Links | BibTeX | Tags: Choice models, Machine learning
@article{Feng2011c,
title = {Assessing the relative importance of input variables for route choice modeling: a neural network approach},
author = {T Feng and T A Arentze and H J P Timmermans},
url = {https://www.jstage.jst.go.jp/article/easts/9/0/9_0_341/_article},
doi = {https://doi.org/10.11175/easts.9.341},
issn = {1881-1124},
year = {2011},
date = {2011-01-01},
urldate = {2011-01-01},
journal = {Journal of the Eastern Asia Society for Transportation Studies},
volume = {9},
pages = {341-353},
abstract = {This paper identifies the relative importance of variables influencing route choice using a neural network approach. Variables related to route attributes and choice contexts are simultaneously incorporated into the model, and a weight partition algorithm is employed to calculate the strength of influence on route choice decisions. The network is trained and validated using stated preference data. Simulation results show good predictability (97.4% of accuracy) of the neural network model. The relative importance of input variables indicates that road category, pricing, bonus and passing through an urban area are more important. Among all choice contexts, the size of truck is most important, followed by travel time difference and road length. The relative importance identified by the neural network model is consistent with the results of a multinomial logit model, and provide meaningful references for variable selection and model estimation.},
keywords = {Choice models, Machine learning},
pubstate = {published},
tppubtype = {article}
}
This paper identifies the relative importance of variables influencing route choice using a neural network approach. Variables related to route attributes and choice contexts are simultaneously incorporated into the model, and a weight partition algorithm is employed to calculate the strength of influence on route choice decisions. The network is trained and validated using stated preference data. Simulation results show good predictability (97.4% of accuracy) of the neural network model. The relative importance of input variables indicates that road category, pricing, bonus and passing through an urban area are more important. Among all choice contexts, the size of truck is most important, followed by travel time difference and road length. The relative importance identified by the neural network model is consistent with the results of a multinomial logit model, and provide meaningful references for variable selection and model estimation.