Dongwoo Lee
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Dongwoo Lee has a total of 21 co-authors. Here are the top co-authors by number of shared publications:
Sybil Derrible
Department of Civil and Materials Engineering, University of Illinois at Chicago, USA
4 shared publications- Sanghyeon Ko
Department of Transportation Planning, Metropolitan Washington Council of Governments, Washington, DC 20002, USA
3 shared publications - Sybil Derrible
Department of Civil, Materials, and Environmental Engineering, University of Illinois Chicago, Chicago, IL
3 shared publications - Binh Thai Pham
Intelligent Geosystems and Infrastructure Research Group, University of Transport Technology, Hanoi, Vietnam
1 shared publication Chana Joanne Haboucha
Department of Civil and Environmental Engineering, Technion, Israel Institute of Technology
1 shared publication- Francisco Camara Pereira
Technical University of Denmark Kongens Lyngby Denmark
1 shared publication
Latest Publications
- Open Access
- 20 July 2026
- Published by Elsevier in SSRN Electronic Journal
From Urban Scaling Laws to Learnable Spatial Mechanisms: Universal Urban Downscaling with Meta-Learning Framework
This study proposes the Universal Urban Downscaling Model (UUDM), a novel meta-learning framework for hierarchical spatial disaggregation. Operating on the premise that universal spatial patterns governed by land use underpin all urban indicators, UUDM employs meta-learning to “learn to learn” these fundamental relationships from rich auxiliary geospatial data. This approach is particularly well-suited to the framework, as it enables rapid adaptation to new urban indicators with minimal data while extracting hierarchical spatial features that capture generalizable patterns underlying diverse urban phenomena. Extensive experiments on a large-scale dataset of 518 urban indicators validate our model. With comprehensive cross-validation, UUDM consistently outperforms specialized baselines, establishing state-of-the-art performance for general-purpose downscaling and proving the viability of a single, unified framework. More critically, the model exhibits remarkable zero-shot generalization. When trained on diverse domains, it accurately performs downscaling for tasks from completely unseen, highly dynamic population domains without any fine-tuning. This confirms that the model learns fundamental, transferable principles of urban spatial dynamics rather than task-specific patterns. UUDM, therefore, can provide a robust methodology for fine-grained urban analytics, enabling integrated policy simulations and reducing reliance on costly domain-specific data collection. Our work translates the idea of urban scaling laws by learning a transferable spatial scaling mechanism for intra-urban inference. Rather than estimating indicator-specific relationships, UUDM learns a transferable spatial scaling mechanism that governs how diverse urban phenomena are redistributed across spatial resolutions, offering a new paradigm for developing urban AI that is both generalizable and highly adaptable to different cities and unforeseen future challenges.
- 20 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessFrom Urban Scaling Laws to Learnable Spatial Mechanisms: Universal Urban Downscaling with Meta-Learning Framework
This study proposes the Universal Urban Downscaling Model (UUDM), a novel meta-learning framework for hierarchical spatial disaggregation. Operating on the premise that universal spatial patterns governed by land use underpin all urban indicators, UUDM employs meta-learning to “learn to learn” these fundamental relationships from rich auxiliary geospatial data. This approach is particularly well-suited to the framework, as it enables rapid adaptation to new urban indicators with minimal data while extracting hierarchical spatial features that capture generalizable patterns underlying diverse urban phenomena. Extensive experiments on a large-scale dataset of 518 urban indicators validate our model. With comprehensive cross-validation, UUDM consistently outperforms specialized baselines, establishing state-of-the-art performance for general-purpose downscaling and proving the viability of a single, unified framework. More critically, the model exhibits remarkable zero-shot generalization. When trained on diverse domains, it accurately performs downscaling for tasks from completely unseen, highly dynamic population domains without any fine-tuning. This confirms that the model learns fundamental, transferable principles of urban spatial dynamics rather than task-specific patterns. UUDM, therefore, can provide a robust methodology for fine-grained urban analytics, enabling integrated policy simulations and reducing reliance on costly domain-specific data collection. Our work translates the idea of urban scaling laws by learning a transferable spatial scaling mechanism for intra-urban inference. Rather than estimating indicator-specific relationships, UUDM learns a transferable spatial scaling mechanism that governs how diverse urban phenomena are redistributed across spatial resolutions, offering a new paradigm for developing urban AI that is both generalizable and highly adaptable to different cities and unforeseen future challenges.
- 7 May 2025
- Published by Elsevier in Accident Analysis & Prevention
- Vol. 218, 108088
Spatiotemporal urban traffic safety analytical framework by integrating nonparametric approaches
- 7 May 2025
- Published by Elsevier in Accident Analysis & Prevention
- Vol. 218, 108088
Spatiotemporal urban traffic safety analytical framework by integrating nonparametric approaches
- Open Access
- 7 January 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (2) , 501
Identifying the Roadway Infrastructure Factors Affecting Road Accidents Using Interpretable Machine Learning and Data Augmentation
In modern society, vehicle accidents have been a factor that has adversely affected national development for a long time. Many countries have tried to solve this issue, and various solutions have been studied. This study aims to design a process for analyzing vehicle accidents to support safety interventions. In the data preprocessing section, a resampling technique was used to solve the data imbalance problem. Then, we applied five different machine learning models for classification by applying hyperparameter optimization. After classification, model-agnostic interpretation techniques were used to interpret the results of a series of machine learning models. Through the above series of processes, we were able to design a process that analyzes vehicle accident data and derives the factors that affect the accident. The classification model that uses XGBoost with ENN (Edited Nearest Neighbor) shows almost 84.3% accuracy. As a result, for “Length” and “Volume”, we found that certain points (Length: 200 m, 29,233 veh/day) were more likely to have an accident. Moreover, variables, such as volume or the volume of heavy vehicle, the probability of an accident increases as the value increases, but in the case of “Lane width” and “Shoulder width”, it can be confirmed that the probability of occurrence decreases as the value increases. These interpretations have meaningful information that could suggest policy recommendations for reducing traffic accidents and can be helpful in establishing effective traffic accident countermeasures.
- 7 January 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (2) , 501
Open AccessIdentifying the Roadway Infrastructure Factors Affecting Road Accidents Using Interpretable Machine Learning and Data Augmentation
In modern society, vehicle accidents have been a factor that has adversely affected national development for a long time. Many countries have tried to solve this issue, and various solutions have been studied. This study aims to design a process for analyzing vehicle accidents to support safety interventions. In the data preprocessing section, a resampling technique was used to solve the data imbalance problem. Then, we applied five different machine learning models for classification by applying hyperparameter optimization. After classification, model-agnostic interpretation techniques were used to interpret the results of a series of machine learning models. Through the above series of processes, we were able to design a process that analyzes vehicle accident data and derives the factors that affect the accident. The classification model that uses XGBoost with ENN (Edited Nearest Neighbor) shows almost 84.3% accuracy. As a result, for “Length” and “Volume”, we found that certain points (Length: 200 m, 29,233 veh/day) were more likely to have an accident. Moreover, variables, such as volume or the volume of heavy vehicle, the probability of an accident increases as the value increases, but in the case of “Lane width” and “Shoulder width”, it can be confirmed that the probability of occurrence decreases as the value increases. These interpretations have meaningful information that could suggest policy recommendations for reducing traffic accidents and can be helpful in establishing effective traffic accident countermeasures.
- Open Access
Exploring the Spatiotemporal Heterogeneities in Urban Vitality Through Scalable Proxies from Mobile Data
Jane Jacob’s concepts of urban vitality and diversity have become prevailing urban planning philosophies in most countries for making cities more livable. Recent changes in demographics and the impacts of COVID-19 have exacerbated the economic and social challenges that cities commonly face, particularly spatiotemporal heterogeneities. Being able to understand these heterogeneities in scalable approaches is fundamental to tackling these challenges in cities. Therefore, this article aims to provide a new form of scalable estimation of urban vitality by using the de facto population. Instead of merely adopting static statistical information such as morphological characteristics in areas, we leverage dynamic factors such as internal mobility flows and energy use intensity as proxies for the spatiotemporal dynamics of indoor and outdoor behaviors of crowds. In this way, we combine dynamic attributes and static features to describe the patterns of urban vitality, which are directly related to spatiotemporal dynamics in urban places. We utilize GNSS-based mobile data and building energy usage intensity as dynamic proxies along with static data such as land use mix and age distribution. To better capture spatial heterogeneity, we use a Multiscale Geographically Weighted Regression (MGWR) model to identify the relationships between the de facto population and the dynamic and static factors. Drawing from the factors determining urban vitality, this article provides policy implications for alleviating spatiotemporal urban imbalances. These data-driven implications can fill the technical knowledge gaps in establishing planning strategies for achieving urban sustainability while enhancing overall subjective livability.
Open AccessExploring the Spatiotemporal Heterogeneities in Urban Vitality Through Scalable Proxies from Mobile Data
Jane Jacob’s concepts of urban vitality and diversity have become prevailing urban planning philosophies in most countries for making cities more livable. Recent changes in demographics and the impacts of COVID-19 have exacerbated the economic and social challenges that cities commonly face, particularly spatiotemporal heterogeneities. Being able to understand these heterogeneities in scalable approaches is fundamental to tackling these challenges in cities. Therefore, this article aims to provide a new form of scalable estimation of urban vitality by using the de facto population. Instead of merely adopting static statistical information such as morphological characteristics in areas, we leverage dynamic factors such as internal mobility flows and energy use intensity as proxies for the spatiotemporal dynamics of indoor and outdoor behaviors of crowds. In this way, we combine dynamic attributes and static features to describe the patterns of urban vitality, which are directly related to spatiotemporal dynamics in urban places. We utilize GNSS-based mobile data and building energy usage intensity as dynamic proxies along with static data such as land use mix and age distribution. To better capture spatial heterogeneity, we use a Multiscale Geographically Weighted Regression (MGWR) model to identify the relationships between the de facto population and the dynamic and static factors. Drawing from the factors determining urban vitality, this article provides policy implications for alleviating spatiotemporal urban imbalances. These data-driven implications can fill the technical knowledge gaps in establishing planning strategies for achieving urban sustainability while enhancing overall subjective livability.
- Open Access
- 30 April 2023
- Published by Korean Society of Transportation in Journal of Korean Society of Transportation
- Vol. 41 (2) , 254-265
Estimation of the Effect of Commercial Vehicle Driving Safety Training Using the Propensity Score Matching Method
교통분야에서 교통안전에 대한 문제는 매우 중요한 이슈가 되었다. 1997년 스웨덴에서 시작한 비전제로 교통안전 정책이 알려진 이후로, 여러 국가에서 도로 교통안전 문제를 해결하기 위해 다양한 정책과 관련 기술이 채택되어지고 있다. 교통안전 정책의 실효성을 높이기 위해서는 이러한 정책과 처치가 교통사고 예방에 미치는 잠재적 영향에 대한 평가가 필요하다. 그러나 자료의 한계와 방법론적 문제로 인해 안전정책의 효과를 정량화 하는데 한계가 있다. 기존 연구들은 통계적 타당성 검증, 경험적 베이지안 방법 등을 주로 채택하였지만 이러한 방법론은 교란요인을 제거하여 교통안전정책의 순수 효과를 측정하는데 한계가 있다. 교통안전정책의 순수효과를 알아보기 위해 본 논문은 교란요인을 통제하는 성향점수매칭방법을 사용하여 교통안전체험교육 정책의 순수효과를 제시하였다. 분석결과 교통안전 체험교육을 이수한 운전자는 이수하지 않은 운전자에 비해서 교통사고 재발율이 16.4% 감소하는 것으로 나타났다. 연령별로 세분화하여 비교한 결과 70대가 36.7%, 60대가 28.5%로 고연령대가 저연령대에 비해 효과가 큰 것으로 나타났으며, 업종별로는 택시 업종이 39.2% 가장 효과가 높은 것으로 나타났다.
- 30 April 2023
- Published by Korean Society of Transportation in Journal of Korean Society of Transportation
- Vol. 41 (2) , 254-265
Open AccessEstimation of the Effect of Commercial Vehicle Driving Safety Training Using the Propensity Score Matching Method
교통분야에서 교통안전에 대한 문제는 매우 중요한 이슈가 되었다. 1997년 스웨덴에서 시작한 비전제로 교통안전 정책이 알려진 이후로, 여러 국가에서 도로 교통안전 문제를 해결하기 위해 다양한 정책과 관련 기술이 채택되어지고 있다. 교통안전 정책의 실효성을 높이기 위해서는 이러한 정책과 처치가 교통사고 예방에 미치는 잠재적 영향에 대한 평가가 필요하다. 그러나 자료의 한계와 방법론적 문제로 인해 안전정책의 효과를 정량화 하는데 한계가 있다. 기존 연구들은 통계적 타당성 검증, 경험적 베이지안 방법 등을 주로 채택하였지만 이러한 방법론은 교란요인을 제거하여 교통안전정책의 순수 효과를 측정하는데 한계가 있다. 교통안전정책의 순수효과를 알아보기 위해 본 논문은 교란요인을 통제하는 성향점수매칭방법을 사용하여 교통안전체험교육 정책의 순수효과를 제시하였다. 분석결과 교통안전 체험교육을 이수한 운전자는 이수하지 않은 운전자에 비해서 교통사고 재발율이 16.4% 감소하는 것으로 나타났다. 연령별로 세분화하여 비교한 결과 70대가 36.7%, 60대가 28.5%로 고연령대가 저연령대에 비해 효과가 큰 것으로 나타났으며, 업종별로는 택시 업종이 39.2% 가장 효과가 높은 것으로 나타났다.
- 22 October 2021
- Published by Taylor & Francis in Transportation Planning and Technology
- Vol. 44 (8) , 843-859
Analyzing travel behavior in Hanoi using Support Vector Machine
This study investigates travel decisions (i.e. travel mode and destination) in Hanoi (Vietnam) using Support Vector Machine (SVM). First, a travel interview survey was conducted and 311 responses were collected across Hanoi. Second, a SVM model was trained to predict travel decisions and compared with a multinomial logit (MNL) model (as a benchmark). Third, the most important variables that affect travel decisions were ranked and discussed. The results show that SVM achieves an accuracy of 76.1% (compared to 72.9% for MNL). Moreover, proposed parking charges, household income, trip mode, and trip cost are found to be the most important variables. In contrast, trip purpose, gender, and occupation are found to negatively affect the model. Overall, low travel cost and low motorcycle parking charges, especially for commuters and shoppers, make people less willing to switch to more sustainable modes such as public and active transport.
- 22 October 2021
- Published by Taylor & Francis in Transportation Planning and Technology
- Vol. 44 (8) , 843-859
Analyzing travel behavior in Hanoi using Support Vector Machine
This study investigates travel decisions (i.e. travel mode and destination) in Hanoi (Vietnam) using Support Vector Machine (SVM). First, a travel interview survey was conducted and 311 responses were collected across Hanoi. Second, a SVM model was trained to predict travel decisions and compared with a multinomial logit (MNL) model (as a benchmark). Third, the most important variables that affect travel decisions were ranked and discussed. The results show that SVM achieves an accuracy of 76.1% (compared to 72.9% for MNL). Moreover, proposed parking charges, household income, trip mode, and trip cost are found to be the most important variables. In contrast, trip purpose, gender, and occupation are found to negatively affect the model. Overall, low travel cost and low motorcycle parking charges, especially for commuters and shoppers, make people less willing to switch to more sustainable modes such as public and active transport.
- Open Access
- 3 September 2021
- Published by MDPI AG in Sustainability
- Vol. 13 (17) , 9910
Interdependencies of Urban Behavioral Dynamics Whilst COVID-19 Spread
The outbreak of novel coronavirus disease 2019 (COVID-19) caused many consequences in almost all aspects of our lives. The pandemic dramatically changes people’s behavior in urban areas and transportation systems. Many studies have attempted to analyze spatial behavior and to present analysis data visually in the process of spreading COVID-19 and provided limited temporal and geographical perspectives. In this article, the behavioral changes in urban areas and transportation systems were analyzed throughout the U.S.A. while the COVID-19 spread over 2020. Specifically, assuming the characteristics are not repetitive over time, temporal phases were proposed where spikes or surges of confirmed cases are noticed. The interdependencies between population, mobility, and additional behavioral data were explored at the county level by adopting the machine learning approaches. As a result, interdependencies with the COVID-19 cases were identified differently by phase. It appeared to have a solid relationship with population size at all phases. Furthermore, it revealed racial characteristics, residential types, and vehicle mile traveled ratio in the urban and rural areas had a relationship with confirmed cases with different importance by phase. Although other short-term analyses were also conducted in terms of the COVID-19, this article is considered more legitimate as it provides dynamic relationships of urban elements by Phase at the county level. Moreover, it is expected to be encouraging and beneficial in terms of phase-driven transportation policy preparedness against a possible forthcoming pandemic crisis.
- 3 September 2021
- Published by MDPI AG in Sustainability
- Vol. 13 (17) , 9910
Open AccessInterdependencies of Urban Behavioral Dynamics Whilst COVID-19 Spread
The outbreak of novel coronavirus disease 2019 (COVID-19) caused many consequences in almost all aspects of our lives. The pandemic dramatically changes people’s behavior in urban areas and transportation systems. Many studies have attempted to analyze spatial behavior and to present analysis data visually in the process of spreading COVID-19 and provided limited temporal and geographical perspectives. In this article, the behavioral changes in urban areas and transportation systems were analyzed throughout the U.S.A. while the COVID-19 spread over 2020. Specifically, assuming the characteristics are not repetitive over time, temporal phases were proposed where spikes or surges of confirmed cases are noticed. The interdependencies between population, mobility, and additional behavioral data were explored at the county level by adopting the machine learning approaches. As a result, interdependencies with the COVID-19 cases were identified differently by phase. It appeared to have a solid relationship with population size at all phases. Furthermore, it revealed racial characteristics, residential types, and vehicle mile traveled ratio in the urban and rural areas had a relationship with confirmed cases with different importance by phase. Although other short-term analyses were also conducted in terms of the COVID-19, this article is considered more legitimate as it provides dynamic relationships of urban elements by Phase at the county level. Moreover, it is expected to be encouraging and beneficial in terms of phase-driven transportation policy preparedness against a possible forthcoming pandemic crisis.
- Open Access
- 31 August 2021
- Published by MDPI AG in Sustainability
- Vol. 13 (17) , 9620
Effectiveness of Green Infrastructure Location Based on a Social Well-Being Index
Urban Green Infrastructure (GI) provides promising opportunities to address today’s pressing issues in cities, mainly resulting from uncurbed urbanization. GI has the potential to make significant contributions to make cities more sustainable by satisfying the growing appetite for higher standards of living as well as helping cities adapt to extreme climate events. To leverage the potentials of GI, this article aims to investigate the effectiveness of GI that can enhance social welfare benefits in the triple-bottom line of urban sustainability. First, publicly available data sets representing social demographic, climate, and built environmental elements are collected and indexed to normalize its different scales by the elements, which is termed as the “Social Well-being Index.” Second, a random forest regressor was applied to identify the impacts of variables on the indexed scores by region. As a result, both the Seoul and Gyeonggi-do models found the most significant relationship with the type of GI to prevent pollutants and disasters, followed by GI types to conserve and improve the environment in Seoul and GI types to serve activity spaces in Gyeonggi-do. Furthermore, variables such as population, number of pollutants, and employment rate in Seoul were found significant and employment rate, population, and air pollution were significant in Gyeonggi-do. Finally, a scenario analysis is conducted to investigate the impacts of the overall index score with additional GI facilitation according to the model’s findings. This article can provide effective strategies for implementing policies about GI by considering regional conditions. The analytical processes in this article can provide useful insights into preparing effective ecological and environmental improvement policies accordingly.
- 31 August 2021
- Published by MDPI AG in Sustainability
- Vol. 13 (17) , 9620
Open AccessEffectiveness of Green Infrastructure Location Based on a Social Well-Being Index
Urban Green Infrastructure (GI) provides promising opportunities to address today’s pressing issues in cities, mainly resulting from uncurbed urbanization. GI has the potential to make significant contributions to make cities more sustainable by satisfying the growing appetite for higher standards of living as well as helping cities adapt to extreme climate events. To leverage the potentials of GI, this article aims to investigate the effectiveness of GI that can enhance social welfare benefits in the triple-bottom line of urban sustainability. First, publicly available data sets representing social demographic, climate, and built environmental elements are collected and indexed to normalize its different scales by the elements, which is termed as the “Social Well-being Index.” Second, a random forest regressor was applied to identify the impacts of variables on the indexed scores by region. As a result, both the Seoul and Gyeonggi-do models found the most significant relationship with the type of GI to prevent pollutants and disasters, followed by GI types to conserve and improve the environment in Seoul and GI types to serve activity spaces in Gyeonggi-do. Furthermore, variables such as population, number of pollutants, and employment rate in Seoul were found significant and employment rate, population, and air pollution were significant in Gyeonggi-do. Finally, a scenario analysis is conducted to investigate the impacts of the overall index score with additional GI facilitation according to the model’s findings. This article can provide effective strategies for implementing policies about GI by considering regional conditions. The analytical processes in this article can provide useful insights into preparing effective ecological and environmental improvement policies accordingly.
- Open Access
- 1 June 2021
- Published by Elsevier in Transportation Research Interdisciplinary Perspectives
- Vol. 10, 100379
Development of a machine learning-based sketch planning model for predicting mobile emissions
- 1 June 2021
- Published by Elsevier in Transportation Research Interdisciplinary Perspectives
- Vol. 10, 100379
Open AccessDevelopment of a machine learning-based sketch planning model for predicting mobile emissions
- 1 January 2020
- Published by American Society of Civil Engineers (ASCE) in Journal of Water Resources Planning and Management
- Vol. 146 (1)
Predicting Residential Water Demand with Machine-Based Statistical Learning
Predicting residential water demand is challenging because of two technical questions: (1) which data and variables should be used and (2) which modeling technique is most appropriate for high prediction accuracy. To address these issues, this article investigates 12 statistical techniques, including parametric models and machine learning (ML) models, to predict daily household water use. In addition, two data scenarios are adopted, one with only 6 variables, generally available to cities and water utilities (general scenario), and one with all 19 variables available from the Residential End-Use 2016 database (REU 2016 scenario). The results for the REU 2016 scenario indicate that ML models outperform linear models. In particular, gradient boosting regression (GBR) performs best with an of 0.69 compared to 0.54 for linear regression. The performance gap between ML and linear models becomes even wider for the general scenario with an of 0.60 for GBR compared to 0.33 for linear regression. The finding in this article can be useful to researchers, municipalities, and utilities seeking novel modeling techniques that can provide consistent modeling performance—i.e., high prediction accuracy—depending on data availability. Future work could include the development of new measures to increase the interpretability of ML models to better understand causal relationships between independent variables and daily household water use.- 1 January 2020
- Published by American Society of Civil Engineers (ASCE) in Journal of Water Resources Planning and Management
- Vol. 146 (1)
Predicting Residential Water Demand with Machine-Based Statistical Learning
Predicting residential water demand is challenging because of two technical questions: (1) which data and variables should be used and (2) which modeling technique is most appropriate for high prediction accuracy. To address these issues, this article investigates 12 statistical techniques, including parametric models and machine learning (ML) models, to predict daily household water use. In addition, two data scenarios are adopted, one with only 6 variables, generally available to cities and water utilities (general scenario), and one with all 19 variables available from the Residential End-Use 2016 database (REU 2016 scenario). The results for the REU 2016 scenario indicate that ML models outperform linear models. In particular, gradient boosting regression (GBR) performs best with an of 0.69 compared to 0.54 for linear regression. The performance gap between ML and linear models becomes even wider for the general scenario with an of 0.60 for GBR compared to 0.33 for linear regression. The finding in this article can be useful to researchers, municipalities, and utilities seeking novel modeling techniques that can provide consistent modeling performance—i.e., high prediction accuracy—depending on data availability. Future work could include the development of new measures to increase the interpretability of ML models to better understand causal relationships between independent variables and daily household water use.