Street network typology defines the structural envelope within which traffic signal control can operate. Planning-stage assessments often treat urban form, demand allocation, signal timing, and microsimulation as separate tasks, leaving an unresolved question: whether coordinated signal timing can offset, or only mitigate, the operational disadvantage associated with an unfavorable street layout. This study develops a planning-stage digital-twin-oriented virtual-city microsimulation framework for evaluating street network typology and coordinated signal control as a coupled infrastructure-control system. Here, digital-twin-oriented virtual-city microsimulation framework denotes an auditable what-if workflow based on a controlled virtual representation, not real-time synchronization, physical-system mirroring, or closed-loop operational control. Four typologies, namely, Traditional Grid, Traditional Superblock, Neo-Traditional Grid, and Neo-Traditional Superblock, are constructed in a common 240-ha virtual city under common land use, population, roadway hierarchy, zoning, demand logic, route-choice treatment, and timing-generation protocol. Signalized intersections are selected using warrant-based criteria, coordinated timings are generated with TRANSYT 7F, and the final performance is evaluated in PARAMICS. The results provide deterministic descriptive evidence that structural-package differences remain after a common timing-generation protocol. Under both center-attraction settings, the superblock package produces a 22-s higher mean stop delay than the corresponding grid package, while reducing the center-attraction share from 46% to 32% lowers the mean stop delay by 6 s within both structural families. Relative to the Traditional Superblock Network, the Neo-Traditional Grid Network reduces mean stop delay from 52 to 24 s and total stop delay from 301,184 to 123,552 s, whereas mean travel distance decreases by only 2.8%. The dominant residual penalty of coarse block structure is therefore interruption accumulation rather than geometric route extension. The contribution is a reproducible planning-stage protocol that converts microsimulation outputs into auditable evidence for topology-sensitive signal coordination and street-network design. The findings should be interpreted as deterministic structural-package contrasts rather than stochastic significance tests, single-factor causal estimates, or real-time operational digital-twin results.
Unobserved heterogeneity is a key factor leading to biased estimates and unreliable crash predictions, undermining the effectiveness of safety countermeasures. Two common options to address it are separating crash types and leveraging advanced statistical models. While finite mixture models have recently demonstrated superiority in this regard, they fail to account for correlations between different crash types. This study thus integrates the two approaches by developing bivariate finite mixture (BFM) models while decomposing crashes into single‐vehicle (SV) and multivehicle (MV) crashes so as to simultaneously address the correlation feature and unobserved heterogeneity. Two variants of the BFM model are developed with SV and MV crash counts as the bivariate response variable. The first variant, BFM‐I‐2, measures the correlation using multivariate normally distributed random error terms, while the second, BFM‐II‐2, employs a joint probability density function. Both bivariate models categorize SV and MV crashes into two subgroups, respectively. Among these models, BFM‐I‐2 exhibits better performance than BFM‐II‐2; it not only verifies the positive correlations between SV and MV crashes but also highlights substantial differences in the key contributing factors across the two crash types. Traffic volume exerts opposing effects on the two subgroups of SV crashes, indicating that unobserved heterogeneity may explain the inconsistent findings regarding its influence in previous research. For MV crashes, the two subgroups share few common risk factors: those in Component 1 are predominantly associated with traffic flow conditions, whereas Component 2 is largely affected by road geometry. These findings can support developing tailored freeway safety interventions.
Police enforcement constitutes a key element of traffic safety management efforts, yet existing studies have largely relied on system-wide and temporally aggregated analyses, limiting insight into spatial heterogeneity and dynamic enforcement effects. This study develops an integrated framework to examine the dynamic relationship between police enforcement and crashes at fine spatial and temporal resolutions, with the objective of characterizing spatial heterogeneity, temporal dynamic, and persistence of enforcement effects. The framework integrates resolution diagnostics, dynamic time-series models, and deep-learning verification. Effective sample size and impulse-response stability metrics are used to identify spatial and temporal resolutions that balance analytical precision with inferential reliability. Vector autoregression and structural vector autoregression models with impulse response functions quantify contemporaneous and lagged enforcement effects, while deep-learning models, including encoder–decoder architectures, transformers, and graph neural networks, provide independent directional verification under flexible nonlinear specifications. The results identified a consistently short-lived inverse citation–crash relationship across all segments, with substantial spatial heterogeneity and generally stronger enforcement responses in urban segments than in rural segments. The consistency between deep-learning sensitivity measures and impulse response estimates supports the robustness of these findings. The results provide actionable and practical insights into spatially and temporally targeted enforcement strategies, supporting data-driven police resource allocation to improve freeway safety.
In recent years, reinforcement learning (RL) has emerged as a powerful platform for developing autonomous driving systems. Nevertheless, while RL policies perform well under ideal conditions, they frequently exhibit unstable behavior when subjected to noisy observations or imperfect actuation factors that significantly compromise safety in real-world deployment. This study seeks to address these vulnerabilities by proposing distribution-level regularization (DLR). Rather than altering the underlying policy architecture, DLR introduces a training constraint that encourages consistent action distributions across both clean and perturbed inputs, facilitating straightforward integration into existing frameworks. The utility of this approach was tested by incorporating DLR into an on-policy proximal policy optimization (PPO) model across discrete- and continuous-action spaces, and further extending it to an off-policy soft actor-critic (SAC) model under continuous control, evaluating its performance across diverse robustness scenarios, including stochastic noise and actuator faults, in highway and intersection environments. The results provide evidence that DLR effectively minimizes collision frequency and extends stable driving duration without sacrificing performance under nominal conditions. These findings suggest that robust autonomous behavior may be achieved more efficiently through output-level constraints rather than through the development of increasingly complex architectures.
Emerging multilayer rail networks in metropolitan areas, comprising multiple service types and suppliers, pose new spatial challenges for operation and governance due to underlying cooperative and competitive dynamics. These challenges increase the need to diagnose the spatial distribution and intensity of interactions among coexisting services. Yet existing interpretation of competitive–cooperative dynamics and available diagnostic methodologies remain insufficient for addressing operational and governance challenges. To address this gap, by extending the concept of coopetition to multilayer rail networks, this study reveals that the substitutability and complementarity embedded in network spatial configurations shape the potential of coopetitive relations between operators. Based on this perspective, we reformulate the problem of measuring networkwide coopetition potential as pairwise relations between rail lines and propose two novel indicators, the network substitutability index (NSI) and the network complementarity index (NCI), to quantify the spatial potential for competition and cooperation between rail lines. To compute these indicators, a bipartite graph–based framework is developed to explicitly represent interline station substitutions and connections, together with an algorithm for efficiently identifying and aggregating relevant origin–destination (OD) relations. In the empirical application of metro and intercity rail systems in the Pearl River Delta (PRD) of Guangdong Province, China, the indicators are computed for all relevant line pairs to enable a systematic, networkwide diagnosis of substitutability and complementarity patterns. Empirical results demonstrate the framework’s applicability and reveal distinct spatial patterns of substitutability and complementarity. These patterns provide a basis for identifying corridors where coordination, differentiation, or regulatory attention may be prioritized, thereby informing service planning and integration‐oriented governance in multilayer railway systems.
Intelligent transportation systems (ITS) have developed into an interdisciplinary research domain shaped by advances in sensing, communications, traffic modeling, data analytics, and artificial intelligence (AI). This study provides a domain‐based bibliometric analysis of ITS research indexed in the Web of Science Core Collection. The main dataset was constructed from a WoS‐only and domain‐based search using the query TS = (“Intelligent Transportation System ∗ ”), which retrieved publications explicitly indexed with the term “Intelligent Transportation System(s).” After applying the SCIE index, document‐type, and year filters, the final core dataset included 7813 publications from 1993 to 2024. Bibliometrix/Biblioshiny, VOSviewer, and complementary visualization tools were used to examine annual publication trends, disciplinary distribution, source journals, country and institutional patterns, citation‐based intellectual foundations, and keyword evolution. To address the potential limitation of the single‐query design, an expanded ITS‐related sensitivity dataset was also constructed using broader terms such as smart transportation, smart mobility, connected vehicles, autonomous driving, traffic prediction, vehicular networks, V2X, Internet of Vehicles, transportation cyber‐physical systems, and AI‐based traffic management. The results show that WoS‐indexed ITS research experienced rapid growth, with an average annual growth rate of 25.86% in the core dataset. China, the United States, and India were the most productive countries in the main dataset, while IEEE Transactions on Intelligent Transportation Systems, IEEE Access, Sensors, and IEEE Transactions on Vehicular Technology were major publication outlets. Citation analyses indicate that traffic flow forecasting, data‐driven ITS, autonomous driving, connected vehicles, and Internet of Vehicles research form important intellectual foundations of the field. Keyword evolution suggests a gradual shift within the WoS‐indexed ITS corpus from infrastructure‐oriented and route‐planning topics toward traffic modeling, intelligent control, connected mobility, and emerging AI‐related themes such as deep learning, graph neural networks, federated learning, autonomous driving, and transformers. The expanded sensitivity dataset contained 27,261 records and showed broadly consistent temporal, geographic, source‐level, and thematic patterns. These results support the robustness of the main findings while confirming that the core dataset should be interpreted as an explicit ITS‐domain corpus rather than as a comprehensive census of all AI‐in‐transportation research. Therefore, the AI‐related findings reported in this study should be understood as themes emerging within WoS‐indexed ITS publications, not as a complete map of AI applications in transportation.
Traffic sign detection (TSD) is a fundamental component of intelligent transportation systems (ITS) and autonomous driving. Despite recent advancements in deep learning, robust TSD in real‐world scenarios remains challenging due to the prevalence of small targets, complex environmental interference (e.g., adverse weather and varying illumination), and the strict requirement for real‐time inference. To address these issues, this paper proposes GAHP‐YOLOv8n, an efficient and robust TSD algorithm. First, a global attention mechanism (GAM) is integrated into the backbone network to enhance the extraction of critical features while suppressing background noise across both channel and spatial dimensions. Second, a specialized small object detection layer is introduced to construct a more effective multiscale feature pyramid, mitigating feature loss for distant targets. Finally, a novel lightweight feature extraction module, C2f‐PKI, is designed to optimize multiscale representation and global context awareness while significantly reducing computational overhead. Experimental results demonstrate that GAHP‐YOLOv8n achieves 80.5% mAP@0.5 and 61.2% mAP@0.5:0.95. This represents a significant improvement of 7.2% and 4.9% over the baseline YOLOv8n, respectively, while maintaining a high inference speed of 189.2 FPS. Comprehensive comparisons validate that the proposed model outperforms current mainstream detectors (e.g., YOLOv9t and YOLOv10n), achieving an optimal trade‐off between accuracy and real‐time efficiency in complex traffic scenarios.
Vehicle trajectory extraction from unmanned aerial vehicle (UAV) aerial videos offers valuable data for traffic safety analysis, especially in high-risk areas such as expressway merge areas. However, the unique characteristics of aerial footage, such as small object sizes, limited distinguishing features, and frequent tracking ID switches, pose significant challenges to accurate detection and tracking. This study proposes a robust framework to address these challenges and enhance vehicle trajectory reconstruction. The YOLOv5 detection model is integrated with a dedicated small-target prediction head and attention mechanisms to improve detection precision under complex backgrounds. To boost tracking accuracy and efficiency, the DeepSORT algorithm is modified by replacing its feature extraction backbone with a lightweight MobileNetV2 network. The resulting trajectories are further refined through filtering and transformed into ground coordinates. A tree-based reconstruction algorithm is applied to enhance temporal continuity. Experimental results show notable improvements: vehicle detection accuracy increased by 8.18%, 8.96%, and 12.62% at aerial altitudes of 200, 250, and 350 m, respectively; multiobject tracking accuracy improved by 6.7%, 8.27%, and 10.9% at the same altitudes. These outcomes demonstrate the effectiveness of the proposed framework for UAV-based traffic analysis in complex environments such as expressway merge areas.
This paper draws on comparative urbanism (CU) and mobility justice (MJ) to examine how contrasting governance and infrastructural regimes shape public transport outcomes in Accra and Geneva. While Geneva’s multimodal transport system reflects institutionalized planning, regulation, and sustainability, Accra’s predominantly informal system demonstrates adaptive urbanism shaped by necessity, flexibility, and everyday negotiation. Rather than viewing formal and informal systems as binary opposites, the paper explores how governance arrangements influence accessibility, affordability, safety, environmental sustainability, and mobility inclusion. Using a structured narrative review of secondary quantitative indicators, policy documents, and institutional reports, the study compares public transport usage, safety outcomes, affordability, environmental performance, and governance capacity across the two cities. The findings reveal substantial disparities. Public transport accounts for 60%–65% of daily trips in Geneva, compared to 40%–45% in Accra, where informal modes constitute over 70% of passenger journeys. Road traffic fatality rates exceed 17 per 100,000 population in Accra, compared to fewer than 3 per 100,000 in Geneva. The study argues that mobility outcomes are shaped not simply by economic development but by the interaction of governance arrangements, institutional capacity, public investment, and policy priorities. By integrating CU and MJ, it highlights how transport systems produce distinct patterns of inclusion, exclusion, opportunity, and risk. The paper concludes that hybrid, context-sensitive approaches combining Accra’s adaptability with Geneva’s institutional coherence offer pathways toward more equitable and sustainable urban mobility.
Undersea tunnels play a crucial role in modern transportation systems by facilitating undersea connectivity and supporting urban mobility. However, the entrance sections of these tunnels pose significant traffic safety challenges, primarily because drivers have limited ability to adjust speed in response to sudden environmental and geometric changes, leading to increased accident risks. Although variable speed limit signs (VSLSs) and visual speed reduction markings (VSRMs) have been proposed to address this issue, existing regulations lack detailed guidance on their coordinated layout. To fill this gap, this study systematically investigates the synergistic effects of combining VSLSs and VSRMs at undersea tunnel entrances. Using UC-win/Road and SketchUp to construct simulation scenarios and employing the Blue Tiger driving simulator to collect behavioral data, 12 coordinated layout schemes were designed, involving different types of deceleration markings (transverse, longitudinal, and fishbone shaped) and various spacing intervals. The comparative analysis of data under each scheme enabled a quantitative evaluation of their effects on speed management and operational stability. The results indicate that fishbone-shaped VSRMs with a 40-m coordination distance achieve the optimal deceleration effect and can be recommended as an optimized layout scheme for improving traffic safety at undersea tunnel entrances.
With the rapid expansion of high-speed railways, ensuring the operational safety of drivers has become a critical challenge for the sustainable development of the transportation sector. However, existing assessments of driver suitability focus mainly on long-term attributes, while systematic frameworks for evaluating drivers’ predeparture suitability status remain underdeveloped. To address this gap, this study screened and validated key indicators from practical contexts through grounded theory and semistructured interviews to construct a multilevel evaluation index system consisting of three dimensions, seven first-level indicators, and fourteen second-level indicators. A combined weighting method integrating the analytic hierarchy process (AHP) and the entropy method was employed to determine indicator weights, thereby balancing expert judgment with objective data. Regression analysis based on data from 123 railway drivers showed that both the composite suitability score and its three dimensions significantly predicted safety performance, with cognitive ability exerting the strongest effect. These findings validate the scientific robustness and practical utility of the proposed framework in diagnosing drivers’ readiness for duty. The study contributes to the literature by bridging the gap between long-term qualifications and real-time safety performance. It further provides railway enterprises and policymakers with a reliable tool to enhance fitness-for-duty screening and to strengthen prediction–intervention–prevention mechanisms in safety management.
Addressing the common oversight of vehicle type differences and their interactions in current traffic flow prediction—particularly the lack of quantitative modeling of passage patterns at toll stations—this paper proposes the spatiotemporal and traffic interaction network (ST-TINet). The model employs a dual-branch architecture to separately analyze the operational characteristics of passenger vehicles and trucks. It integrates key neighboring station passage data through a spatiotemporal attention module and innovatively designs a passenger–freight interaction module to quantify competitive behaviors between vehicle types. Ultimately, it achieves joint prediction of traffic flow and electronic toll collection (ETC) usage rates. Empirical studies based on three typical highway toll stations in Shandong Province demonstrate that the proposed method significantly outperforms traditional baseline approaches in both prediction accuracy and stability, especially at stations with large traffic flow and complex traffic patterns. The framework’s dynamic ETC demand estimation and vehicle-type-level flow prediction provide data support for dynamic lane allocation at toll stations, truck off-peak travel guidance, and differentiated toll rate policy formulation, thereby empowering the refined development of intelligent transportation systems.
Vehicle trajectory prediction is a critical capability for safe autonomous driving, enabling vehicles to anticipate the motions of surrounding road users and plan collision-free maneuvers. However, this task remains highly challenging due to complex spatiotemporal dependencies, dynamic multiagent interactions, and the inherently multimodal and uncertain nature of human driving behavior. Recent work increasingly adopted Transformer architectures, leveraging self-attention mechanisms to model long-range temporal dependencies and interaction effects. This paper presents a focused literature review of Transformer-based methods for vehicle trajectory prediction. Following a systematic selection workflow, we searched Scopus, IEEE Xplore, and Google Scholar using trajectory-prediction and attention-related keywords and applied a two-stage screening process to identify relevant studies. The reviewed set spans 2020–2025 and is used to address three research questions concerning architectural patterns, comparative performance relative to earlier model families, and deployment barriers. The review addresses three core research questions: (1) architectural patterns and innovations in Transformer applications, (2) comparative performance against earlier model families, and (3) practical deployment barriers. Furthermore, we synthesize Transformer design trends, including encoder–decoder and encoder-only variants, hierarchical and multistage attention for spatiotemporal and social modeling, hybrid integrations with graph neural networks (GNNs), convolutional neural networks (CNNs), and recurrent neural networks (RNNs), as well as emerging large pretrained foundation models on trajectory data. We further summarize common input representations and scene encoding practices, approaches for modeling interactions, and multimodal prediction strategies. Finally, we review datasets and evaluation metrics commonly used in the literature and discuss open challenges such as robustness to noisy or incomplete inputs, model interpretability for safety validation, and computational efficiency for real-time on-board deployment.
Hurricane Irma stands as one of the most destructive tropical storms to make landfall in the United States, particularly impacting the State of Florida, where it prompted the largest evacuation in history with approximately 7 million residents. The profound consequences of mass evacuation underscore the critical need to understand travel behaviors during hurricane evacuation and the recovery process. This research analyzes statewide evacuation and re-entry patterns, leveraging diverse datasets, including TTMS data from main corridors and GIS data. A statewide corridor-based empirical analysis framework is constructed to characterize evacuation and re-entry response patterns using sensor-based traffic observations. The results show that evacuation and re-entry curves can characterize corridor-level emergency response and recovery patterns. The observed evacuation response varied across corridors and was associated with the spatial structure of the evacuation network, storm progression, and the distribution of major evacuation routes. Re-entry showed weaker and less consistent corridor-level spatial heterogeneity than evacuation and, in several locations, displayed stepwise or intermittent recovery patterns. In addition, the I-10 corridor exhibited a distinct east–west interstate pattern, with prolonged westbound evacuation loading and relatively concentrated eastbound re-entry traffic. The findings provide empirical insights for state agencies in emergency transportation management, corridor planning, and postdisaster recovery assessment.
With the acceleration of urbanization and rapid growth of vehicle ownership, traffic congestion has become a critical bottleneck constraining sustainable urban development. Traditional traffic signal control methods struggle to adapt to the complex traffic environment where connected and automated vehicles (CAVs) coexist with conventional vehicles. This perspective review synthesizes current research advances with practical implementation insights to address these challenges, examining multi‐intersection cooperative control platform frameworks for CAV environments from both theoretical and engineering perspectives. Current platform developments typically adopt three‐layer architecture designs comprising a data access layer, a data middleware layer, and a data application layer, achieving decoupling between data management platforms and algorithm integration platforms, thereby enhancing system flexibility and scalability. In terms of data management, a graph‐structured data model is introduced to better represent road network topology, and a multilevel data fusion mechanism is established at the parameter, feature, and decision levels. Regarding functional implementation, three core functions are integrated: intelligent speed guidance, emergency vehicle signal priority, and blind spot safety warning. Through the collaborative operation of on‐board units, roadside units, and control centers, these platforms enable organic integration of “smart vehicles” and “intelligent roads.” The platform design follows the core philosophy of “perceiving congestion, identifying congestion, and alleviating congestion.” Case study results from implementations in urban areas, such as Guangzhou’s central district, demonstrate potential for traffic efficiency improvements, with reported delay reductions of approximately 10% at key intersections. In the intelligent connected transportation system with electronic toll collection test in Foshan, vehicle identification accuracy exceeded 99.9%, with all system functions and performance indicators meeting design requirements. These developments provide insights into engineering‐implementable solutions for traffic control in intelligent connected environments, offering theoretical perspectives and practical implications for advancing smart transportation development. Through this integrated perspective that combines structured narrative analysis with validated implementation frameworks, this work aims to bridge the persistent gap between theoretical innovation and engineering practice in CAV cooperative control systems.
In the past decades, there has been a significant growth in awareness related to the impacts of school commuting in metropolitan areas. The use of public transport and the promotion of active transportation choices among educational communities are key issues for the development of sustainable and efficient cities. This paper explores multiple factors, which range from the socioeconomic sphere to the built environment and safety around schools and also includes variables such as the proximity of the school to public transport stations/stops, the school grade and the school type (i.e., public or private schools). This paper aims to study how all these factors influence two different school commuting choices in Lisbon, Portugal: (1) the choice between active versus motorised travel modes and (2) the choice between public versus private transport modes. The Hands Up survey database, which includes 10 different modes for school commuting, was analysed to estimate binomial logistic regression models. The results suggest that school commuting independence happens at younger ages for public school students than for private school students and that a higher socioeconomic status has a negative impact regarding the use of public transport and active commuting to school. The planning of the public transport network is also pivotal since the close proximity of stations/stops of certain commuting modes discourages active commuting to school.
Within the competitive landscape of civil aviation, achieving precision in decision-making is crucial for enhancing enterprise competitiveness and revenue. This study addresses the integrated aircraft routing problem by combining flight scheduling, fleet assignment, and routing problems into an integrated framework. The stochastic attribute of flight delays poses significant challenges to decision-making, necessitating consideration within the decision process. This research proposes an integrated aircraft routing model, incorporating real-world constraints and features such as optional flights and buffer times to absorb stochastic delay impacts. A novel column generation algorithm is employed to efficiently solve the model, which includes a greedy optimal insertion algorithm for initializing decision variables and a labeling-setting algorithm for subproblem exploration. The study emphasizes the importance of accurately setting flight buffer times to mitigate stochastic delay risks. Numerical experiments on a benchmark competition dataset show that the proposed algorithm improves the initial solution by about 19% within 10.67 s and yields solutions within 10% of the MILP optimum. Sensitivity analysis further indicates that increasing the pool size of the labeling-setting algorithm from 100 to 2000 raises the objective value by about 6.1% while increasing CPU time from 2.55 to over 300 s, highlighting the trade-off between schedule quality and computational effort in practical airline applications. This research contributes to both theoretical understanding and practical optimization of aircraft routing in dynamic operational environments, holding significant implications for civil aviation management.
Elderly drivers represent a growing and vulnerable population on U.S. roadways, exhibiting distinct collision patterns and heightened risk of severe injury. Accurate prediction and interpretation of collision types involving this demographic are critical for developing targeted safety interventions. This study evaluates the performance of four machine learning models-Kolmogorov-Arnold Networks (KANs), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Traditional Neural Networks (TNNs)-in classifying the manner of collision among elderly drivers using data from the Fatality Analysis Reporting System (FARS). KANs offer a novel, interpretable neural network architecture that can capture complex nonlinear relationships with explicit feature representations, addressing the interpretability challenge common in deep learning. Model performance is assessed through accuracy, balanced accuracy, macro and weighted F1-scores, and confusion matrices, while SHapley Additive exPlanations (SHAP) provide insight into feature importance and interaction effects. Results indicate that KANs achieve competitive predictive performance comparable to RF, XGBoost, and TNNs, with the added advantage of enhanced interpretability. SHAP analysis reveals key predictors and feature interactions influencing the predicted collision type, such as the combined effect of speed limit and roadway surface condition shifting predictions toward angle and head-on configurations and the interaction of lighting and weather influencing the model's distinction across collision types. These findings demonstrate the potential of interpretable machine learning approaches to improve understanding of crash configurations among elderly drivers, enabling more informed policy and engineering interventions.
Low-altitude corridor development is accelerating across Asia, raising practical questions about how emerging unmanned and urban air mobility operations may interact with conventional hub airport performance once sustained corridor activity becomes observable. This study constructs a preintegration baseline assessment for 18 major Asian hub airports using monthly data from 2019 to 2022. Public ADS-B records from the OpenSky Network are combined with NOAA Global Summary of the Day weather aggregates and airport structural descriptors to derive three airport-month indicators: arrival-spacing dispersion, short-gap share at or below 120 s and relative runway-use intensity benchmarked to each airport’s 2019 peak arrival rate. Because sustained, credibly coded corridor activation is not observable within the study window, corridor treatment effects are not estimated. Two-way fixed-effects models show that higher monthly movements are associated with lower log-transformed spacing dispersion (β = −2.06 × 10−4, p<0.001) and a higher short-gap share (β = 1.30 × 10−5, p<0.001). In the relative runway-use intensity model, more adverse weather is associated with lower runway-use intensity for the reference group, while high-density and CDM interaction terms are not statistically robust. Robustness checks using raw and winsorised variance, lower-tail spacing, pandemic-period exclusion, 2019-only estimation and removal of sparse-coverage hubs preserve the negative movement coefficient for spacing dispersion. The study contributes a reproducible measurement pipeline and strategic monitoring framework for corridor-readiness screening before sustained low-altitude corridor integration, rather than a causal evaluation of corridor effects.
The growing complexity of car to two-wheeler accident patterns underscores the urgent need for sustainable solutions in traffic safety research. Current limitations in analyzing key variables and predicting accident severity hinder the effectiveness of prevention strategies, highlighting a critical gap in achieving sustainability goals. Traditional LightGBM approaches, while valuable, require extensive hyperparameter tuning and often suffer from overfitting, compromising model accuracy and long-term sustainability. To address these challenges, this study proposes an improved gray wolf optimization (IGWO)-optimized LightGBM classification model, integrating sustainability principles to enhance accident severity prediction. The IGWO algorithm employs a dynamic weight allocation mechanism for fitness coefficients and a nonlinear step-size update strategy, significantly improving convergence precision and reducing susceptibility to local optima, key factors for sustainable model performance. Experimental results demonstrate IGWO's superior convergence accuracy compared with baseline GWO (p < 0.05), validating its potential for sustainable optimization. The IGWO-LightGBM model achieves balanced parameter tuning, ensuring robustness and reproducibility, cornerstones of sustainable research. Shapley Additive exPlanations (SHAP) analysis identifies the collision type as the most influential feature (SHAP value: 1.5, 42 percent higher than others), guiding targeted interventions. Environmental and spatiotemporal analyses identify critical sustainability risks; road accidents with higher severity levels occur when adhesion coefficients drop below 0.4 during winter midnight periods. These findings advocate for prioritizing high-conflict scenarios, low-adhesion road maintenance, and temporal risk management to foster sustainable traffic safety.
