Subham Kharel

Bridges are critical to the US transportation network and are essential components of our infrastructure systems. Coastal floods can disrupt these structures, isolating communities and highlighting the urgent need for effective maintenance and prioritization strategies. This study introduces a digital twin framework that evaluates the impacts of coastal floods on bridge substructures by integrating multisource data and employing spatial algorithms like inverse distance weighting and k - d tree methods to model flood severity. Using a generative adversarial network (GAN) with focal loss, the framework simulates preflood and postflood scenarios, generating synthetic data sets that reflect realistic damage patterns and evolving maintenance priorities. The GAN-enhanced model improved F1 scores (the harmonic mean of precision and recall, providing a balance between the two metrics), particularly for Condition 4 (bridges with advanced section loss, deterioration, spalling, or scour), where the score increased from 0.63 to 0.87 preflood and from 0.63 to 0.88 postflood, effectively addressing class imbalances. It also enhanced macro and weighted average F1 scores, demonstrating robust predictive capabilities. Confusion matrices showed increased identification of bridges in Conditions 4 and 5, consistent with the model’s emphasis on vulnerable infrastructure. The study’s findings reveal shifts in maintenance priorities when flood severity is incorporated, emphasizing bridges with higher susceptibility to combined flooding and socioeconomic impacts. While the framework primarily focused on physical infrastructure impacts, it also highlighted potential disparities in resource allocation, suggesting the need for further research into equitable distribution strategies. The digital twin provides actionable insights for local and federal agencies, offering a valuable tool for managing and prioritizing infrastructure maintenance in response to extreme weather events while laying the groundwork for more inclusive resource planning.

This paper introduces a hybrid Artificial Intelligence framework for continuous prediction on im balanced, high-dimensional tabular data, demonstrated through travel mode choice prediction in the Lehigh Valley region as a case study. Traditional classification approaches struggle when dominant classes overshadow minority classes, leading to biased predictions and poor minority-class perfor mance. To address this, the framework integrates Boruta feature selection to identify informative predictors, Generative Adversarial Networks (GANs) to generate synthetic samples while empha sizing hard-to-learn observations, and eXtreme Gradient Boosting (XGBoost) regression to predict continuous outcomes, capturing complex nonlinear interactions among features. Evaluated on cen sus block group–level socioeconomic, land-use, and travel network data, the approach substantially improves prediction accuracy for continuous outcomes while maintaining interpretability through feature importance quantified using SHapley Additive exPlanations (SHAP). Beyond transporta tion, this framework provides a generalizable methodology for AI-driven continuous prediction in high-dimensional, imbalanced tabular datasets, offering actionable insights for diverse real-world applications.

Accurately modeling traffic flows in large urban networks remains challenging, even for classical equilibrium and modern AI-based approaches, due to complex spatial dependencies, adaptive route choice, and multi-path behavior. This paper presents a novel AI-driven traffic assignment framework that integrates Graph Neural Networks (GNNs) to learn network-wide flow patterns from a representative sample of origin-destination (OD) pairs, Proximal Policy Optimization (PPO) to model adaptive multi-path routing, and classical congestion-based updates to ensure realistic travel times. By probabilistically distributing flows across multiple competitive paths and dynamically incorporating congestion feedback, the framework predicts traffic patterns for unobserved OD pairs without exhaustive path enumeration. Empirical validation on real-world networks demonstrates that the proposed method achieves scalable, accurate, and behaviorally realistic traffic assignment, effectively capturing congestion dynamics and traveler decision-making in complex transportation systems, while reducing computation time from days to minutes.

This paper introduces a hybrid Artificial Intelligence framework for continuous prediction on im balanced, high-dimensional tabular data, demonstrated through travel mode choice prediction in the Lehigh Valley region as a case study. Traditional classification approaches struggle when dominant classes overshadow minority classes, leading to biased predictions and poor minority-class perfor mance. To address this, the framework integrates Boruta feature selection to identify informative predictors, Generative Adversarial Networks (GANs) to generate synthetic samples while empha sizing hard-to-learn observations, and eXtreme Gradient Boosting (XGBoost) regression to predict continuous outcomes, capturing complex nonlinear interactions among features. Evaluated on cen sus block group–level socioeconomic, land-use, and travel network data, the approach substantially improves prediction accuracy for continuous outcomes while maintaining interpretability through feature importance quantified using SHapley Additive exPlanations (SHAP). Beyond transporta tion, this framework provides a generalizable methodology for AI-driven continuous prediction in high-dimensional, imbalanced tabular datasets, offering actionable insights for diverse real-world applications.

Off-system bridges are critical components of the United States’ transportation infrastructure, providing essential access to rural communities and enabling residents to reach vital services such as employment, education, and healthcare. Many of these bridges are structurally deficient, functionally obsolete, and unmaintained. This disproportionately hinders the mobility of underserved populations, worsening socioeconomic disparities. Despite existing research, there is insufficient focus on the unique challenges posed by off-system bridges, including handling the class imbalanced nature of the bridge condition rating dataset. This paper predicts bridge deck conditions by using Generative Adversarial Networks with Focal Loss (GAN-FL) to generate synthetic data which enhances precision–recall balance in imbalanced datasets. Results show that integrating GAN-FL with random forest (RF) classifiers significantly enhances the performance of minority classes, improving their precision, recall, and F1 scores. The study finds that augmenting training data using GAN-FL greatly enhances minority class prediction, thereby assisting in accurate bridge condition modeling.

Additive manufacturing (AM) or 3D printing is transforming the construction industry by enabling the production of complex structures and components from digital blueprints using materials like concrete, plastics, and recycled materials. This technology reduces material waste, lowers production costs, and opens new possibilities for sustainable and affordable housing. Traditionally used for prototypes and low-volume production, AM has advanced into the architecture, engineering, and construction (AEC) sectors, offering potential solutions to the affordable housing crisis. Concrete 3D printing, for example, can reduce carbon emissions through the use of alternative materials, minimizing the need for raw resources. Additionally, the ability to optimize material usage and reduce construction waste through techniques like prefabrication and rapid construction can significantly lower the cost of building homes. This paper discusses how AM can contribute to addressing the challenges of affordable housing by exploring its applications in construction, its potential for reducing environmental impacts, and its role in improving cost-effectiveness. By integrating AM into manufactured housing models, it becomes possible to develop sustainable, cost-effective homes on a larger scale, which offers a promising solution to the growing demand for affordable housing. Through the widespread adoption of 3D printing technologies, it is feasible to address both affordability and sustainability concerns in the housing sector.

Accessibility to food stores offering nutritious and affordable options is critical to ensure urban residents lead healthy lives. However, the uneven distribution of diverse food options in U.S. cities has raised concerns about marginalized communities’ ability to obtain affordable and nutritious food. Transportation-related issues, like car-oriented urban form and sprawl, have created food deserts in cities, making it hard for marginalized communities to access affordable and healthy food. This lack of access has recently drawn the attention of urban planners to address food access-related equity issues. Spatial interaction models can help measure food access by examining how grocery store supply, demand, and proximity affect individuals’ access to necessary food. However, these measures do not always consider the pricing of food items, which can significantly affect food access for low-income individuals and families. We use a highly disaggregated measure tailored to measuring inequitable food accessibility by factoring grocery store pricing into various types of supply and demand. Then, using regional socio-demographic thresholds, we identify target population groups and perform a series of equity analyses using our food accessibility measure. Our results show that walking is more inequitable than driving to food stores and that African American and Hispanic minorities, the elderly, and carless households are more burdened than the reference groups (whites). Using our framework, we identify food deserts and provide spatial and empirical insights for policymakers and planners to address food insecurity and promote equitable access to healthy food.

Auto-centric streets and zoning regulations that have segregated land use in the U.S.A. has created an inequitable and less accessible transportation system, particularly for vulnerable communities. We examined whether the existence or absence of zoning regulations influenced transportation equity outcomes, by focusing on case studies of Dallas, a city with zoning regulations, and Houston, a city without zoning regulations. While many studies use single-modal measures to examine modal accessibility gaps (MAGs), we adopted a multi-modal accessibility measure. This approach allowed us to calculate white- and blue-collar job accessibilities for both transit and automobiles, providing a nuanced estimate of MAGs. Our maps show considerable variations in MAGs, with Dallas being more equitable than Houston. Results from the spatial regression models show that (a) white- and blue-collar job seekers face similar levels of MAG inequalities, (b) MAGs decreases when transitioning from Dallas to Houston, suggesting that Dallas offers higher transit accessibility equity than Houston, (c) commercial land uses have considerably higher spillover impacts (than direct impacts) on MAGs, (d) unlike Houston, where there are negative associations between commercial land use and MAGs, Dallas shows more robust positive associations, suggesting the potential for greater equity among its residents, (e) the spillover impact of single-family housing units on MAGs in Houston is higher than its direct impact, and (f) transit proximity has a larger direct impact but smaller spillover effects on transportation equity. Planners should implement comprehensive long- and short-term strategies favoring communities of concern by considering land use and transportation decisions together to address socio-economic inequalities.