The National Transportation Safety Board (NTSB) database contains investigator-written narratives that summarize what happened and why in civil aviation accidents and incidents. While manual reading is effective for individual cases, it does not scale to cross-event mechanism comparison and evidence aggregation, and causal mechanisms are often expressed in free-text rather than structured fields. To support scalable analysis, this study focuses on the narr_cause field as an investigator-curated causal statement and aims to extract a structured causal backbone from these narratives. We propose ARCEM (Aviation Report Causality Extraction Model), which formulates causality extraction as a word-level sequence labeling task and identifies Cause, Effect, and Embedded-causality spans. ARCEM integrates a foundation-model contextual embedding module (BERT), a Transformer-based multi-granular word embedding module that fuses trainable word embeddings with lightweight character-morphology features extracted by TranCCNN, and a Transformer-based causal inference module with a CRF decoder. Experiments on manually annotated NTSB narr_cause sentences show that ARCEM outperforms strong baselines (including SCITE, EMGCE, and CharBERT), achieving 0.8961 accuracy, 0.8515 precision, 0.8425 recall, and 0.8470 F1-score. Error and ablation analyses verify the contributions of key components, and a causal-cue masking stress test indicates that ARCEM’s gains are not solely driven by recognizing overt cue phrases. Finally, we provide a qualitative case study that converts word-level predictions into compact causality-chain visualizations, offering a practical representation for downstream safety analytics.

Traffic safety remains a critical priority within urban transportation systems, and real-time safety analysis is essential for timely risk identification and proactive intervention. Most existing studies on conflict prediction have examined the effects of traffic conditions, vehicle status, and geometric factors on conflict risk. However, driver behaviors, especially conflict evasive behaviors, also play a crucial role. To address this gap, this study analyzed vehicle evasive patterns during rear-end conflicts. An improved K-Means clustering algorithm was applied to identify six evasive patterns under both high- and low-risk conflict conditions. Key influencing factors included roadway and traffic characteristics, following- and leading-vehicle attributes, and non-conflict driving behaviors under normal driving conditions. Following variable selection, a two-level Nested Logit (NL) model was established to jointly predict evasive patterns and conflict risk levels, achieving predictive accuracy above 0.70 and AUC greater than 0.90. Key findings revealed that braking behavior under non-conflict conditions was strongly correlated with subsequent evasive responses, whereas congested traffic conditions tended to delay driver reactions. Leading-vehicle characteristics, such as speed, acceleration, and headway, significantly influenced following-vehicle responses. These findings suggest that incorporating evasive driving patterns into conflict risk prediction can provide a more comprehensive and individualized understanding of rear-end conflicts on expressways.

Previous studies in predicting crash risks primarily associated the number or likelihood of crashes on a road segment with traffic parameters or geometric characteristics, usually neglecting the impact of vehicles' continuous movement and interactions with nearby vehicles. Recent technology advances, such as Connected and Automated Vehicles (CAVs) and drones, are able to collect high-resolution trajectory data, which enable trajectory-based risk analysis. This study investigates a new vehicle group (VG) based risk analysis method and explores risk evolution mechanisms considering VG features. An impact-based vehicle grouping method is proposed to cluster vehicles into VGs by evaluating their responses to the erratic behaviors of nearby vehicles. The risk of a VG is aggregated based on the risk between each vehicle pair in the VG, measured by inverse Time-to-Collision (iTTC). Logistic Regression and a Graph Neural Network (GNN) are used to predict VG risks based on both aggregated and disaggregated VG information. Both methods achieve excellent performance with AUC values exceeding 0.93. For the GNN model, GNNExplainer with feature perturbation is applied to identify critical individual vehicle features and their directional impact on VG risks. Overall, this research contributes a new perspective for identifying, predicting, and interpreting traffic risks.

With increasing demands for precision in traffic control, evaluation, and other trajectory-based applications, reconstructing sparse trajectories has become a prominent topic. This study presents an innovative approach employing the particlike filter (PF) algorithm to directly output reconstructed second-by-second trajectories at signalized intersections, grounded in Markov assumptions. Instead of allocating travel time, this method incorporates the influence of signal status for trajectory reconstruction. By integrating historical trajectory data with real-time signal status, the current vehicle motion is categorized to estimate the expected position for the next second. These serve as referential data for generating particles and weight assignments in the PF algorithm. Numerical experiments using the NGSIM dataset confirmed the accuracy and stability of this approach through analysis of average error, comparisons with other methods, and parametric sensitivity analysis. The reconstruction accuracy is improved by 6.37% when the model processing is classified based on signal status.

Reinforcement Learning (RL) is widely used in tasks where agents interact with an environment to maximize rewards. Building on this foundation, Safe Reinforcement Learning (Safe RL) incorporates a cost metric alongside the reward metric, ensuring that agents adhere to safety constraints during decision-making. In this paper, we identify that Safe RL is vulnerable to backdoor attacks, which can manipulate agents into performing unsafe actions. First, we introduce the relevant concepts and evaluation metrics for backdoor attacks in Safe RL. It is the first attack framework in the Safe RL field that involves both Positive and Negative Action sample (PNAct) is to implant backdoors, where positive action samples provide reference actions and negative action samples indicate actions to be avoided. We theoretically point out the properties of PNAct and design an attack algorithm. Finally, we conduct experiments to evaluate the effectiveness of our proposed backdoor attack framework, evaluating it with the established metrics. This paper highlights the potential risks associated with Safe RL and underscores the feasibility of such attacks. Our code and supplementary material are available at https://github.com/azure-123/PNAct.

Accurate and timely detection of incident-induced congestion (IIC) is essential for mitigating its negative impact on traffic efficiency. Existing studies on IIC detection mainly focus on traffic flow on freeways and face challenges on urban roads due to the impacts of signal lights at intersections and diverse road networks. Additionally, the low penetration rate of probe vehicle trajectories poses another challenge. This study proposes a probe-vehicle-trajectory-based algorithm for IIC detection on urban roads at the movement and cycle levels. Two critical features (i.e., the average speed and the entrance time into the road segment) are defined to capture the characteristics of trajectory segments. A Vehicle Trajectory Polar Coordinate Transformation (VTPCT) method is proposed to differentiate anomalous trajectory segments (ATS) affected by IIC from normal ones, considering the periodicity of fixed signal timing at the intersections. Anomaly rates calculated from the identified ATS within a spatiotemporal window are introduced to reflect the movement-cycle-level traffic states. A two-stage algorithm framework is designed to enhance the algorithm’s adaptability to spatially sparse trajectories and diverse road networks. Experimental studies show that the proposed algorithm is applicable to trajectory data with low penetration rates and outperforms benchmarks of typical statistical and AI-based algorithms.

ObjectivesThere have been numerous studies on conflict risk for expressways, with the majority of previous studies focusing on road-segments' conflict risk while neglecting the impact of moving vehicles. In recent years, some studies have begun to work on vehicle-group without the consideration of the traffic on the road-segment. However, as the vehicle-group travels along the road-segment, the conflict risk of road-segment and vehicle-group will interact with each other. The aim of this study is to analyze the interactive mechanism of conflict risk between vehicle-groups and road-segments on expressways.MethodsThis study utilized high-resolution vehicle trajectory data to separately build conflict risk prediction models for vehicle-groups and road-segments. The best performing models were selected and explainability algorithms were applied. The analysis then focused on two aspects: (1) the influence of downstream high-risk vehicle-groups on upstream road-segment conflict risk and (2) the impact of geometric and traffic parameters of downstream road-segments on the conflict risk of vehicle-groups.ResultsThe results show that vehicle-group characteristics significantly affect road-segment conflict risk. When high-risk vehicle-groups appear downstream, the conflict risk of the road-segment increases by about 6%, and the impact is stronger when the propagation distance is shorter. In turn, when the conflict risk of a downstream road-segment increases, this risk propagates upstream through the traffic flow, affecting the behavior of vehicle-groups and raising their conflict risk. A difference of 29% in vehicle-group conflict risk was observed depending on the median downstream road-segment conflict risk.ConclusionsThis study demonstrates that vehicle-groups and road-segments interact in the propagation of expressway conflict risk. Integrating these two dimensions enables more accurate conflict risk prediction and analysis. This research provides a novel perspective by integrating vehicle-group and road-segment interactions for more accurate conflict risk prediction and analysis.

Lane-changing maneuvers are crucial driving behaviors closely linked to various collisions, such as rear-end and sideswipe collisions. Precisely predicting lane-changing maneuvers can aid drivers in making informed decisions, thus enhancing driving safety. However, existing research primarily focuses on successful lane-changing maneuvers, neglecting failed ones. This study comprehensively investigates the detection and prediction of failed lane-changing maneuvers in discretionary scenarios using naturalistic vehicle trajectory data. The Mexican hat wavelet (MHW) is employed to accurately detect key time points in successful or failed lane-changing events, including the start, occurrence of failure, and end of the lane-changing maneuver. Subsequently, a failed lane-changing prediction model based on GA-XGBoost is developed to proactively perceive whether a lane-changing maneuver will succeed before it initiates. Moreover, the Shapley Additive exPlanations (SHAP) technique assesses feature importance and interaction effects between features in the prediction model. The results demonstrate that MHW effectively identifies critical time points in lane-changing events. The proposed GA-XGBoost model achieves an impressive 94.55% accuracy in predicting failed lane-changing maneuvers before the driver initiates the lane-changing maneuver. SHAP values reveal that failed lane-changing maneuvers often result from a high collision risk between the subject vehicle (SV) and the following vehicle in the target lane (FVT). Consequently, drivers should pay increased attention to approaching vehicles from behind. Moreover, accelerating during lane-changing helps maintain a safe distance from FVT, improving the likelihood of a successful lane-changing. Integrating the proposed model into advanced driver-assistance systems or autonomous driving systems has the potential to significantly enhance driving safety.

The weaving segment is the bottleneck of expressways, as onramp and offramp traffic intermingles. While numerous studies have focused on predicting lane change behaviors within these segments, many existing approaches fail to exploit the synergistic potential of multisource perception data. This study proposes a novel two-stage prediction model that integrates onboard and roadside perception for enhanced lane change prediction within weaving segments. The proposed methodology comprises sequential intention recognition and trajectory prediction stages. In the first stage, the model leverages roadside perception to assist in inferring early lane change intentions through latent traffic information. In the second stage, it predicts precise lane-changing trajectories with the predicted lane change intentions as prior knowledge. The model's performance was validated using field trajectory data. Specifically, the F1-score is improved by 4.77% for the right lane change intention category. The average displacement error (ADE) and final displacement error (FDE) of predicted trajectories are reduced by 15.3% and 10.1%, respectively. This study provides a fresh perspective on lane change prediction within weaving segments and achieves inspiring results.

Aviation accidents, referring to unexpected and undesirable events involving aircraft, often cause great damage to property and human life. Learning from historical accidents is pivotal for improving safety in aviation. However, aviation accidents are typically documented and stored as unstructured or semi-structured free-text,renderingthe ability to analyze such data a difficult task. This study presents a novel framework that combines text mining and network analytics techniques that provide the ability to analyze aviation accident reports automatically. The framework comprises a four-step modelling approach: (1) the transformation of unstructured aviation safety report texts into structured numeric matrices using the TF-IDF matrix; (2) the identification of aviation accident topics using a StructuralTopicModel(STM); (3) the production of a Word Co-occurrence Network (WCN) to determine the interrelations between aviation safety risk factors; and (4)quantitative analysisby technology of keywords to pinpoint key causal factors in aviation safety events. The proposed framework is validated by analyzing aviation accident reports collected by the National Transportation Safety Board (NTSB). The results indicate that STM provides a more granular partitioning of topics and better distinguishes between similar events compared to traditional Latent Dirichlet Allocation (LDA). Among the identified topics, “Fuel and Power” and “En-route Phase” have the highest occurrence rate according to STM. Additionally, “Aircraft Crash” is the most prevalent topic in aviation accidents that resulted in fatal injuries, while “Landing phase” is the most prevalent topic in non-fatal injuries on accidents. Based on the word co-occurrence network, three centrality measures highlight “inspection of equipment” and “take off” as the most important risk factors in aviation safety. The proposed framework provides a comprehensive solution for in-depth analysis of aviation safety reports, offering decision support for aviation safety management and accident prevention, thereby reducing risks and strengthening safety measures.

Chain conflicts would cause chain-reaction crashes, which might result in elevated fatality rates. Chain conflicts describe a phenomenon wherein evasive actions taken by a following vehicle's driver after a conflict impact nearby vehicles, which occur frequently but are reported less often. To effectively reduce conflict risk, comprehending the evolution patterns of chain conflicts under varied traffic conditions and road segments is crucial, in order to make chain conflicts management strategies. Initially, rear-end or sideswipe conflicts between two vehicles are identified based on vehicle trajectory data captured by an unmanned aerial vehicle group. Subsequently, a chain conflict identification algorithm is proposed, considering the randomness of occurrence time and fluctuation of impact duration, to link individual conflicts. Chain conflict rates exhibit significant variations across different road segments under diverse traffic conditions. Multiple risk and propagation indicators are extracted to unveil latent characteristics of chain conflicts from a high-level perspective. Based on prominent characteristic disparities, three evolution patterns are identified, i.e., Longitudinal Risk Decrease Pattern, Longitudinal Risk Increase Pattern, and Comprehensive High-risk Persistent Pattern. Spatial-temporal high-risk areas associated with each pattern are determined, and transition probabilities between patterns are calculated. The results indicate that these patterns tend to remain stable, with transitions mainly occurring from low-risk to high-risk patterns. Moreover, strategies to reduce conflict risk are proposed based on the characteristics of different patterns. This study holds great significance in understanding chain conflict evolution patterns and preventing chain-reaction crashes.

Connected and Autonomous Vehicles (CAVs) might significantly enhance the transportation system by improving safety, accessibility, efficiency, and sustainability. However, a major challenge lies in ensuring CAVs can operate properly under diverse weather conditions, which have already proven to impair human driving capabilities. This pioneering study aims to bridge a crucial research gap by comprehensively assessing the performance of CAVs on traffic operations and safety across varying weather scenarios. Using microscopic traffic simulation in VISSIM and the Surrogate Safety Assessment Model (SSAM), this study evaluates key metrics, including average speed, delay, number of stops, travel time, and number of conflicts for different CAV market penetration rates. The analysis spans 21 scenarios under clear, light rain, heavy rain, and foggy conditions within a selected urban corridor in the United Arab Emirates. The results showed that the average speed rose by 55% in clear weather, while the average delay, the number of stops, travel time, and the number of accidents decreased by 50%, 50%, 95%, and 68%, respectively. In light rain, the average speed improved by 43%, while the average delay, number of stops, travel time, and the number of accidents reduced by 43%, 56%, 96%, and 74%, respectively. The average speed increased by 82% under heavy rain, while the average delay, the number of stops, the travel time, and the number of accidents all fell by 62%, 68%, 96%, and 74%, respectively. In fog, the average speed rose by 32%, while the average delay, average stop number, travel time, and the number of accidents decreased by 33%, 47%, 90%, and 83%, respectively. Overall, this paper highlights the need for resilient CAV systems adaptable to diverse environmental conditions. It helps advance the understanding of how CAVs can be optimized for safety and efficiency in urban settings, contributing to sustainable transportation solutions. It provides insights into the challenges and innovative approaches for CAV deployment in adverse weather, laying a foundation for future research and the broader implementation of these technologies in urban mobility.

The integrated variable speed limits and ramp metering (VSL-RM) strategy is a useful method to avoid crashes and alleviate congestion on expressways. Previous studies have mostly focused on specific roadway segments and optimized them with the single goal of efficiency or safety, which does not allow for a prompt response to high-risk moving vehicle groups to improve safety and efficiency. To reduce the crash and congestion risk of vehicle groups in real time, this study developed three VSL-RM strategies with different optimization objectives based on predicted risks in a mixed traffic flow environment including connected vehicles (CVs) and regular vehicles (RVs). Due to the different behaviors of CVs and RVs under the VSL-RM control strategy, a mixed traffic METANET model was introduced to predict traffic flow parameters, e.g., volume and speed. Furthermore, two risk prediction models were utilized to predict crash risk and congestion risk based on the traffic flow parameters predicted by the mixed-traffic METANET model. The objectives of the three VSL-RM strategies were to minimize the crash risk, congestion risk, and both crash and congestion risks of the vehicle groups, respectively. These strategies were evaluated using a well-calibrated micro-simulation network. The mixed flow METANET model and the risk prediction model were validated to be consistent with the simulated traffic flow. The results demonstrated that the three strategies could simultaneously improve safety and efficiency benefits in most scenarios. However, the safety-targeted strategy provided the highest safety benefits, while the efficiency-targeted strategy provided the highest efficiency benefits. The bi-objective strategy outperformed the other two strategies in balancing the benefits of safety and efficiency. Moreover, increasing the CV penetration rate resulted in higher benefits for all three strategies.

Sparse trajectory data with non-second-by-second sampling intervals are common. However, most carbon emission estimation models for vehicles require second-bysecond inputs. Additionally, some models ignore the emission generation principle, and some have complicated inputs. To address these limitations, this study proposes a vehicle carbon emission estimation method for urban traffic, based on sparse trajectory data. First, a trajectory reconstruction method based on interpolation of acceleration distribution is proposed. The results showed that the reconstructed trajectory was close to the real trajectory, and the accuracy was 2%-17% higher than that of other methods. Second, a carbon emission estimation model that considers both the emission generation principle and feasibility is proposed. The model with a goodness-of-fit of 0.887 had the best performance compared to the other models. The emission estimation results of the reconstructed sparse trajectories showed that the precision improved significantly for data with different frequencies compared to that of other reconstruction methods, e.g., 9% higher at a 30 s sampling interval. (c) 2024 Tongji University and Tongji University Press. Publishing Services by Elsevier B.V. on behalf of KeAi Communications Co., Ltd. This is an open access article under the CC BYNC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

In-lane street hawking is the intermittent entry of signalized intersections by traders to sell groceries to drivers and passengers. Studies have shown that hawkers get exposed to traffic injuries but the lack of quantitative analysis of their lane entry and exit behaviors in signalized intersections makes it difficult to improve traffic safety. This study analyzes the significant predictors of in-lane street hawkers' (1) lane entry within 30 s after the red signal illumination, (2) lane exit within 30 s after the green signal illumination, and (3) probability of getting injuries during the green signal time. Drone-based trajectory data were collected from a selected signalized intersection in Accra, Ghana. A Weibull accelerated failure time duration model incorporating Gamma frailty was used to evaluate hawkers' behaviors. Overall, the majority of hawkers exhibited red-light running behaviors exposing them to traffic injuries. An increase in traffic speed, especially beyond 20 km/h, exposed hawkers to injury risks significantly. Notably, hawkers' lane entry decreased significantly as the traffic speed increased. Their lane exit duration was significantly predicted by the queue lengths and traffic volumes. Accordingly, safety practitioners can enhance traffic regulation and control methods in addition to pro-poor social interventions to demotivate hawking at signalized intersections.

Safety performance functions (SPFs) have become valuable tools for estimating the relationships between crashes and various causal factors when constructing crash-prediction models. However, the commonly used independent variable, the annual average daily traffic (AADT) is data on a yearly basis, which has limitations in capturing the temporal characteristics of traffic flows influenced by the passage of time. Accordingly, there have also been many studies using 15 min data to reflect real-time, which is an important time unit to understand changes in highway traffic flow. However, such a short time unit has the limitation of high instability and randomness. In light of this, this study recognizes the importance of the 15 min time interval and proposes a new approach by developing a modified hourly model that aggregates data at fine-grained 15 min intervals (00, 15, 30, and 45 min, both at the beginning and end), instead of the traditional hourly data that starts and ends at the peak of each hour to compensate for the existing limitations. The analysis focused on South Korea’s nationwide highways, and models were developed based on both statistical and machine-learning approaches to compare their performances for selecting the final model. Additionally, a modified temporal SPF is introduced to predict crashes by assigning weights based on a Dirichlet distribution to models with overlapping time intervals aggregated in 15 min increments. This innovative approach overcomes the limitations of existing 15 min models, where the number of crashes is too small for effective training if the model is simply developed by dividing the time. The anticipated outcome is that the proposed model will demonstrate excellent performance and serve as an effective tool for predicting highway crash risks.

Truck-involved crashes cause serious social and economic losses, and the crash severity is higher compared to ordinary vehicles. Driver fatigue is a major crash causation, especially have a larger impact on truck drivers. Therefore, mandating a truck rest-break is one of the road safety management strategies for reducing both truck-involved crash frequency and severity. This paper analyses the change in safety performance according to the revision of the truck rest-break time on highways. Because spatial heterogeneity and homogeneity should be considered to estimate safety effects precisely after executing crash countermeasures, this study estimates the safety effect after the revision of the Trucking Transport Business Act, by developing a model based on the Multilevel Full-Bayes before and after the study. The safety performance function was developed in different section units and distributions, and the crash modification factor was calculated for the most suitable model. As a result of the analysis, the performance of the Multilevel Bayesian Poisson-gamma model was the highest, and it was found that there was a crash reduction effect after the revision of the truck break time. The results of this study can be referred to when preparing measures for road safety regulations.