A Bayesian multivariate hierarchical spatial joint model for predicting crash counts by crash type at intersections and segments along corridors

A negative binomial Lindley approach considering spatiotemporal effects for modeling traffic crash frequency with excess zeros

Statistical analysis of traffic crash frequency is significant for figuring out the distribution pattern of crashes, predicting the development trend of crashes, formulating traffic crash prevention measures, and improving traffic safety planning systems. In recent years, the theory and practice for traffic safety management have shown that road crash data have characteristics such as spatial correlation, temporal correlation, and excess zeros. If these characteristics are ignored in the modeling process, it may seriously affect the fitting performance and prediction accuracy of traffic crash frequency models and even lead to incorrect conclusions. In this research, traffic crash data from rural two-way two-lane from four counties in Pennsylvania, USA was modeled considering the spatiotemporal effects of crashes. First, a negative binomial Lindley spatiotemporal effect model of crash frequency was constructed at the micro level; Simultaneously, the characteristics and problems of excess zeros and potential heterogeneity of the crash data were resolved; Finally, the effects of road characteristics on crash frequency were analyzed. The results indicate a significant spatial correlation between the crash frequency of adjacent road sections. Compared with the negative binomial model, the negative binomial Lindley model can better handle the excess zeros characteristics in traffic crash data. The model that considers both spatial correlation and temporal conditional autoregressive effects has the best fit for the observed data. In addition, for road sections that allow passing and have a speed limitation of not less than 50 miles per hour, the crash frequency corresponding to these sections is lower due to their good visibility and road conditions. The increase in average turning angle and intersection density on the horizontal curve of the road section corresponds to an increase in crash frequency.

A spatiotemporal deep learning approach for pedestrian crash risk prediction based on POI trip characteristics and pedestrian exposure intensity

Pedestrians represent a population of vulnerable road users who are directly exposed to complex traffic conditions, thereby increasing their risk of injury or fatality. This study first constructed a multidimensional indicator to quantify pedestrian exposure, considering factors such as Point of Interest (POI) attributes, POI intensity, traffic volume, and pedestrian walkability. Following risk interpolation and feature engineering, a comprehensive data source for risk prediction was formed. Finally, based on risk factors, the VT-NET deep learning network model was proposed, integrating the algorithmic characteristics of the VGG16 deep convolutional neural network and the Transformer deep learning network. The model involved training non-temporal features and temporal features separately. The training dataset incorporated features such as weather conditions, exposure intensity, socioeconomic factors, and the built environment. By employing different training methods for different types of causative feature variables, the VT-NET model analyzed changes in risk features and separately trained temporal and non-temporal risk variables. It was used to generate spatiotemporal grid-level predictions of crash risk across four spatiotemporal scales. The performance of the VT-NET model was assessed, revealing its efficacy in predicting pedestrian crash risks across the study area. The results indicated that areas with concentrated crash risks are primarily located in the city center and persist for several hours. These high-risk areas dissipate during the late night and early morning hours. High-risk areas were also found to cluster in the city center; this clustering behavior was more prominent during weekends compared to weekdays and coincided with commercial zones, public spaces, and educational and medical facilities.

Comparison of corridor-level fatal and injury crash models with site-level models for network screening purposes on Florida urban and suburban divided arterials

Objective: Develop corridor-level network screening models to identify high-risk corridors where safety improvements could be implemented to reduce fatal and injury (FI) crashes. Methods: A novel corridor definition focused on context classification and lane count was developed and applied to urban and suburban four-lane divided arterial roadways in Florida. Negative binomial regression models were developed for multi- and single-vehicle crashes using 80% of the corridors (training set). Crash frequency predictions were obtained from the developed corridor models and similar site-level models from the Highway Safety Manual (HSM) models for the remaining 20% of the corridors (testing set). Results from all models were adjusted using the empirical Bayes (EB) method. Results: A total of 130 corridors were identified across seven counties. These corridors contained approximately 349 km (217 miles) of roadway and experienced 11,437 multi-vehicle and 746 single-vehicle crashes that resulted in fatalities or injuries from 2017 to 2021. After applying the HSM site-level models and the developed corridor-level models to the testing set (both with and without EB adjustments), the corridor-level models with EB adjustments were the most accurate for corridor crash prediction. Applying the corridor-level models with EB adjustments to the testing set gave a predicted value of 386.44 crashes/year, which was the closest to the observed crash frequency of 383.20 crashes/year. From the corridor-level models, a 3.48-km (2.16-mile) high-risk corridor in Miami-Dade County was identified and analyzed site-by-site using the HSM methodology to identify specific sites within the corridor where safety improvements could provide the most FI crash reductions. Conclusions: The corridor-level models were more accurate and statistically reliable than similar HSM models while being less data intensive. They also only required corridor-level data rather than data for each intersection and segment. By using readily available data, the methods in this paper can be easily replicated by agencies to develop their own network screening corridor-level models and expedite the identification of corridors in need of safety improvements to reduce FI crashes. Existing site-level network screening methods can be used to supplement the developed corridor-level methodology by identifying high-risk sites within identified high-risk corridors.

A Hierarchical Multinomial Logit model to examine the effects of signal strategies on right-turn crash risks by crash movement configuration

Crash risk models relying on total crash counts are limited in their ability to extract meaningful insights regarding the context of crashes and to identify effective remedial measures. In addition to the typical classification of collisions noted in the literature (e.g., angle, head-on and rear-end), crashes can also be categorised according to vehicle movement configurations (Definitions for Coding Accidents or DCA codes in Australia). This classification presents an opportunity to extract useful insights into road traffic collision causes and contributing factors that are highly contextual. With this aim, this study develops crash-type models by DCA crash movement, with a focus on right-turn crashes (equivalent to left-turn crashes for right-hand traffic) at signalised intersections using a novel approach for linking crashes with signal control strategies. The modelling approach with contextual data enables quantification of the effect of signal control strategies on right-turn crashes, offering potentially unique and novel insights into right-turn crash causes and contributing factors. Crash-type models are estimated with the crash data of 218 signalised intersections in Queensland from 2012 to 2018. Multilevel (Hierarchical) Multinomial Logit Models with random intercepts are employed to capture the hierarchical influence of factors on crashes and unobserved heterogeneities. These models capture upper-level influences on crashes from intersection characteristics and lower-level influences from individual crash characteristics. The models specified in this way account for the correlation among crashes within intersections and influences on crashes across spatial scales. The model results reveal that the probabilities of the opposite approach crash type are significantly higher than the same direction and adjacent approach crash types for all right-turn signal control strategies at intersections except the split approach, for which the opposite is true. The results also suggest that the number of right-turning lanes and occupancy in conflicting lanes are positively associated with the likelihood of crashes for the same direction crash type.

A comparative study of collision types between automated and conventional vehicles using Bayesian probabilistic inferences

INTRODUCTION

Automated vehicle (AV) technology is a promising technology for improving the efficiency of traffic operations and reducing emissions. This technology has the potential to eliminate human error and significantly improve highway safety. However, little is known about AV safety issues due to limited crash data and relatively fewer AVs on the roadways. This study provides a comparative analysis between AVs and conventional vehicles on the factors leading to different types of collisions.

METHOD

A Bayesian Network (BN) fitted using the Markov Chain Monte Carlo (MCMC) was used to achieve the study objective. Four years (2017-2020) of AV and conventional vehicle crash data on California roads were used. The AV crash dataset was acquired from the California Department of Motor Vehicles, while conventional vehicle crashes were obtained from the Transportation Injury Mapping System database. A buffer of 50 feet was used to associate each AV crash and conventional vehicle crash; a total of 127 AV crashes and 865 conventional vehicle crashes were used for analysis.

RESULTS

Our comparative analysis of the associated features suggests that AVs are 43% more likely to be involved in rear-end crashes. Further, AVs are 16% and 27% less likely to be involved in sideswipe/broadside and other types of collisions (head-on, hitting an object, etc.), respectively, when compared to conventional vehicles. The variables associated with the increased likelihood of rear-end collisions for AVs include signalized intersections and lanes with less than 45 mph speed limit.

CONCLUSIONS

Although AVs are found to improve safety on the road in most types of collisions by limiting human error leading to vehicle crashes, the current state of the technology shows that safety aspects still need improvement.

Road environment characteristics and fatal crash injury during the rush and non-rush hour periods in the U.S: Model testing and cluster analysis

This study aims to assess the relationship between county-level fatal crash injuries and road environmental characteristics at all times of the day and during the rush and non-rush hour periods. We merged eleven-year (2010 - 2020) data from the Fatality Analysis Reporting System. The outcome variable was the county-level fatal crash injury counts. The predictor variables were measures of road types, junction types and work zone, and weather types. We tested the predictiveness of two nested negative binomial models and adjudged that a nested spatial negative binomial regression model outperformed the non-spatial negative binomial model. The median county crash mortality rates at all times of the day and during the rush and non-rush hour periods were 18.4, 7.7, and 10.4 per 100,000 population, respectively. Fatal crash injury rate ratios were significantly elevated on interstates and highways at all times of the day - rush and non-rush hour periods inclusive. Intersections, driveways, and ramps on highways were associated with elevated fatal crash injury rate ratios. Clusters of high fatal crash injury rates were observed in counties located in Montana, Nevada, Colorado, Kansas, New Mexico, Oklahoma, Texas, Arkansas, Mississippi, Alabama, Georgia, and Nevada. The built and natural road environment factors are associated with county-level fatal crash injuries during the rush and non-rush hour periods. Understanding the association of road environment characteristics and the cluster distribution of fatal crash injuries may inform areas in need of focused intervention.

Analysis of injuries and deaths from road traffic accidents in Iran: bivariate regression approach

BACKGROUNDS

This study aims to estimate and compare the parameters of some univariate and bivariate count models to identify the factors affecting the number of mortality and the number of injured in road accidents.

METHODS

The accident data used in this study are related to Kermanshah province in march2020 to march2021. Accidents areas were divided into 125 areas based on density characteristics. In a one-year period, 3090 accidents happened on the suburban roads of Kermanshah province, which resulted in 398 deaths and 4805 injuries. Accident information, including longitude and latitude of accident location, type of accident (fatal and injury), number of deaths, number of injuries, accident type, the reason of the accident, and the kind of accident were all included as population-level variables in the regression models. We investigated four frequently used bivariate count regression models for accident data in the literature.

RESULTS

In bivariate analysis, except for the DNM model, there is a reasonable decrease in the AIC measures of the saturated model compared to the reduced model for the other three models. For the injury models, MSE is lowest, respectively for DIBP (137.87), BNB (289.46), BP (412.36) and DNM (3640.89) models. These results are also established for death models. But, in univariate analysis, only injury models almost present reasonable results.

CONCLUSIONS

Our findings show that the IDBP model is better suitable for evaluating accident datasets than other models. Motorcycle accidents, pedestrian accidents, left turn deviance, and dangerous speeding were all significant variables in the IDBP death model, and these parameters were linked to accident mortality.

Predicting and Analyzing Road Traffic Injury Severity Using Boosting-Based Ensemble Learning Models with SHAPley Additive exPlanations

Road traffic accidents are one of the world’s most serious problems, as they result in numerous fatalities and injuries, as well as economic losses each year. Assessing the factors that contribute to the severity of road traffic injuries has proven to be insightful. The findings may contribute to a better understanding of and potential mitigation of the risk of serious injuries associated with crashes. While ensemble learning approaches are capable of establishing complex and non-linear relationships between input risk variables and outcomes for the purpose of injury severity prediction and classification, most of them share a critical limitation: their “black-box” nature. To develop interpretable predictive models for road traffic injury severity, this paper proposes four boosting-based ensemble learning models, namely a novel Natural Gradient Boosting, Adaptive Gradient Boosting, Categorical Gradient Boosting, and Light Gradient Boosting Machine, and uses a recently developed SHapley Additive exPlanations analysis to rank the risk variables and explain the optimal model. Among four models, LightGBM achieved the highest classification accuracy (73.63%), precision (72.61%), and recall (70.09%), F1-scores (70.81%), and AUC (0.71) when tested on 2015–2019 Pakistan’s National Highway N-5 (Peshawar to Rahim Yar Khan Section) accident data. By incorporating the SHapley Additive exPlanations approach, we were able to interpret the model’s estimation results from both global and local perspectives. Following interpretation, it was determined that the Month_of_Year, Cause_of_Accident, Driver_Age and Collision_Type all played a significant role in the estimation process. According to the analysis, young drivers and pedestrians struck by a trailer have a higher risk of suffering fatal injuries. The combination of trailers and passenger vehicles, as well as driver at-fault, hitting pedestrians and rear-end collisions, significantly increases the risk of fatal injuries. This study suggests that combining LightGBM and SHAP has the potential to develop an interpretable model for predicting road traffic injury severity.

Quantifying and comparing the effects of key risk factors on various types of roadway segment crashes with LightGBM and SHAP

Understanding and quantifying the effects of risk factors on crash frequency is of great importance for developing cost-effective safety countermeasures. In this paper, the effects of key crash contributing factors on total crashes and crashes of different collision types are analyzed separately and compared. A novel Machine Learning (ML) method, Light Gradient Boosting Machine (LightGBM), is introduced to model a Texas dataset consisting of vehicle crashes occurred from 2015 to 2017. Compared with other commonly used ML methods such as eXtreme Gradient Boosting (XGBoost), LightGBM performs significantly better in terms of mean absolute error (MAE) and root mean squared error (RMSE). In addition, the SHapley Additive explanation (SHAP) approach is employed to interpret the LightGBM outputs. Significant risk factors are identified, including speed limits, area type, number of lanes, roadway functional class, shoulder width and shoulder type. With the SHAP method, the importance, total effects, and main and interaction effects of risk factors are quantified. The results suggest that the importance of risk factors vary across collision types. Speed limit is a more important risk factor than right/left shoulder width, lane width, and median width for Rear-End (RE) crashes, while the opposite relationship is found for Run-Off-Road (ROR) crashes. Also, it is found that narrow lanes (8ft to 11ft) increase the risk for all types of crashes (i.e., Total, ROR, and RE) in this study. For road segments with 5 or 6 lanes in both directions combined, a lane width greater than or equal to 12ft may help reduce the risk of all types of crashes. These results have important implications for developing accurate crash modification factors and cost-effective safety countermeasures.

A joint probability model for pedestrian crashes at macroscopic level: Roles of environment, traffic, and population characteristics

Road safety is a major public health issue, with road crashes accounting for one-fourth of all documented injuries. In these crashes, pedestrians are more vulnerable to fatal and/or severe injuries than car occupants. Therefore, it is necessary to have a better understanding of the relationship between pedestrian crashes and possible influencing factors, including road environment, traffic conditions, and population characteristics. In conventional studies, separate prediction models were established for pedestrian crashes and other crash types, which could have ignored possible correlations among the different crash types. Additionally, these influencing factors can contribute to pedestrian crashes in two manners, i.e., contributing to crash occurrence and propensity of pedestrian involvement. Furthermore, extensive pedestrian count data were generally not available, affecting the estimation of pedestrian crash exposure. In this study, a joint probability model is adopted for the simultaneous modeling of crash occurrence and pedestrian involvement in crashes; effects of possible influencing factors, including land use, road networks, traffic flow, population demographics and socioeconomics, public transport facilities, and trip attraction attributes, are considered. Additionally, trip generation and pedestrian activity data, based on a comprehensive household travel survey, are used to determine pedestrian crash exposure. Markov chain Monte Carlo full Bayesian approach is then applied to estimate the parameters. Results indicate that crash occurrence is correlated to traffic flow, number of non-signalized intersections, and points of interest such as restaurants and hotels. By contrast, population age, ethnicity, education, household size, road density, and number of public transit stations could affect the propensity of pedestrian involvement in crashes. These findings indicate that better design and planning of built environments are necessary for safe and efficient access for pedestrians and for the long-term improvement of walkability in a high-density city such as Hong Kong.

A Bayesian spatial Poisson-lognormal model to examine pedestrian crash severity at signalized intersections

Reducing nonmotorized crashes requires a profound understanding of the causes and consequences of the crashes at the facility level. Generally, existing literature on bicyclists and pedestrian crash models suffers from two distinct problems: lack of exposure/volume data and inadequacy in capturing potential correlations across various crash aspects. To develop a robust framework for pedestrian crash analysis, this research employed a multivariate model across multiple pedestrian crash severities incorporating a crucial piece of information: pedestrian exposure. A multivariate spatial (conditional autoregressive) Poisson-lognormal model in a Bayesian framework was developed to examine the significant factors influencing the fatal, incapacitating injury (or suspected serious injury), and non-incapacitating injury pedestrian crashes at 409 signalized intersections in the Austin area. Various explanatory variables were used to examine the pedestrian crashes, including traffic characteristics, road geometry, built environment features, and pedestrian exposure volume at intersections, which was estimated through a direct demand model as part of the study. Model results revealed valuable insights. The superior performance of the multivariate model over the univariate model emphasized the need to jointly model multiple pedestrian crash severities. The results showed the significant positive influence of speed limit on fatal pedestrian crashes and revealed that both incapacitating and non-incapacitating injury crashes increase with increasing motorized traffic volume. Bus stop presence was found to have a negative influence on incapacitating injury crashes and a positive influence on non-incapacitating injury crashes. Moreover, the pedestrian volume at intersections positively influences non-incapacitating injury crashes. The difference in influence across crash types warrants careful and focused policy design of intersections to reduce pedestrian crashes of all severity types.

Spatial relationships between alcohol outlet densities and drunk driving crashes: An empirical study of Tianjin in China

INTRODUCTION

Numerous studies have demonstrated the close relationship between alcohol availability and alcohol-related crashes. However, there is still a lack of spatial empirical analysis regarding this relationship, particularly in large cities of developing countries. Differences in alcohol outlets and drinking patterns in these cities may lead to quite different patterns of crash outcomes.

METHOD

3356 alcohol-related crashes were collected from the blood-alcohol test report of a forensic institution in Tianjin, China. Density of alcohol outlets such as retail locations, entertainment venues, restaurants, hotels, and companies were extracted based on 2114 Traffic Analysis Zones (TAZ) together with the residential and demographic characteristics. After applying the exploratory spatial data analysis, this research developed and compared the traditional Ordinary Least Square model (OLS), Spatial Lag Model (SLM), Spatial Error Model (SEM) and Spatial Durbin Model (SDM) to explore spatial effects of all the variables.

RESULTS

The results of incremental spatial autocorrelation show that the most significant distance threshold of alcohol-related roadway traffic crashes is 3 km. The SDM is found to be the optimal spatial model to characterize the relationship between alcohol outlets and crashes. The number of alcohol-involved traffic crashes is positively related to population density and retail density, but negatively related to the company density, hotel density, and residential density within the same TAZ. Meanwhile, dense population and hotels have reverse spillover effects in adjacent zones.

CONCLUSIONS

The significant spatial direct effect and spillover effect of alcohol outlet densities on drunk driving crashes should not be neglected. These findings could help improve transportation planning, traffic law enforcement and traffic management for large cities in developing countries.

The role of striking and struck vehicles in side crashes between vehicles: Bayesian bivariate probit analysis in China

OBJECTIVE

Side crashes between vehicles which usually lead to high casualties and property loss, rank first among total crashes in China. This paper aims to identify the factors associated with injury severity of side crashes at intersections and to provide suggestions for developing countermeasures to mitigate the levels of injuries.

METHOD

In order to investigate the role of striking and struck vehicles in side crashes simultaneously, bivariate probit model was proposed and Bayesian approach was employed to evaluate the model, compared to the corresponding univariate probit model.

DATA

Crash data from Beijing, China for the period 2009-2012 were used to carry out the statistical analysis. Based on the investigation with vehicles and data analysis on events, 130 intersection side crash cases were selected to form a specific dataset. Then, the influence of human, vehicles, roadway and environmental variables on crash severity was examined by means of bivariate probit regression within Bayesian framework.

RESULTS

The effects of the factors on striking vehicle drivers and struck vehicle drivers were considered separately and simultaneously to find more targeted conclusions. The statistical analysis revealed vehicle type, lane number, no non-motorized lane and speeding have the corresponding influence on the injury severity of striking vehicles, while time of day and vehicle type of struck vehicles increased the likelihood of being injured.

CONCLUSIONS

From the results it can be concluded that there indeed exists correlation between striking and struck vehicles in side crashes, although the correlation is not so strong. Importantly, Bayesian bivariate probit model can address the role of striking and struck vehicles in side crashes simultaneously and can accommodate the correlation clearly, which extends the range of univariate probit analysis. The general and empirical countermeasures are presented to improve the safety at intersections.

Spatial analysis of traffic accidents near and between road intersections in a directed linear network

Although most of the literature on traffic safety analysis has been developed over areal zones, there is a growing interest in using the specific road structure of the region under investigation, which is known as a linear network in the field of spatial statistics. The use of linear networks entails several technical complications, ranging from the accurate location of traffic accidents to the definition of covariates at a spatial micro-level. Therefore, the primary goal of this study was to display a detailed analysis of a dataset of traffic accidents recorded in Valencia (Spain), which were located into a linear network representing more than 30 km of urban road structure corresponding to one district of the city. A set of traffic-related covariates was constructed at the road segment level for performing the analysis. Several issues and methodological approaches that are inherent to linear networks have been shown and discussed. In particular, the network was defined in a way that allowed the explicit investigation of traffic accidents around road intersections and the consideration of traffic flow directionality. Zero-inflated negative binomial count models accounting for spatial heterogeneity were used. Traffic safety at road intersections was specifically taken into account in the analysis by considering the higher variability and number of zeros that can be observed at these road entities and the differential contribution of the covariates depending on the proximity of a road intersection. To complement the results obtained from the count models fitted, coldspots and hotspots along the network were also detected, with explanatory objectives. The models confirmed that spatial heterogeneity, overdispersion and the close presence of road intersections explain the accident counts observed in the road network analyzed. Hotspot detection revealed that several covariates whose contribution was unclear in the modelling approaches may also be affecting accident counts at the road segment level.

Mapping the knowledge domain of road safety studies: A scientometric analysis

As a way of obtaining a visual expression of knowledge, mapping knowledge domain (MKD) provides a vision-based analytic approach to scientometric analysis which can be used to reveal an academic community, the structure of its networks, and the dynamic development of a discipline. This study, based on the Science Citation Index Expanded (SCIE) and Social Sciences Citation Index (SSCI) articles on road safety, employs the bibliometric tools VOSviewer and CitNetExplorer to create maps of author co-citation, document co-citation, citation networks, analyze the core authors and classic documents supporting road safety studies and show the citation context and development of such studies. It shows that road safety studies clustered mainly into four groups, whose we will refer to as "effects of driving psychology and behavior on road safety", "causation, frequency and injury severity analysis of road crashes", "epidemiology, assessment and prevention of road traffic injury", and "effects of driver risk factors on driver performance and road safety", respectively. Through our analysis, the core publications and their citation relationships were quickly located and explored, and "crash frequency modeling analysis" has been identified to be the core research topic in road safety studies, with spatial statistical analysis technique emerging as a frontier of this topic.

Modeling correlation and heterogeneity in crash rates by collision types using full bayesian random parameters multivariate Tobit model

Crashes present different collision types. There usually exist unobserved risk factors which could jointly affect crash rates of different types, resulting in correlation and heterogeneity issues across observations. The primary objective of the study is to propose a novel random parameters multivariate Tobit (RPMV-Tobit) model for evaluating risk factors on crash rates of different collision types. Crash data from 367 freeway diverge areas in a three-year period were obtained for modeling. Three major types of collisions including rear-end, sideswipe, and angle collisions were considered. The RPMV-Tobit model was structured to simultaneously accommodate correlations between crash rates across collision types and unobserved heterogeneity across observations. The RPMV-Tobit model was compared with a multivariate Tobit (MV-Tobit) model, a random effect multivariate Tobit (REMV-Tobit) model, and independent univariate Tobit (IU-Tobit) models under the Bayesian framework. The results showed that MV-Tobit model outperforms the IU-Tobit models on fitting crash rates, indicating that accounting for the correlation between crash types can improve model fit. The RPMV-Tobit model and REMV-Tobit model perform better than the MV-Tobit model, suggesting that accounting for the unobserved heterogeneous can further improve model fit. The improvement of model performance with the RPMV-Tobit model is higher than that with the REMV-Tobit model. The impacts of each risk factor on crash rates were estimated and some differences were found across different collision types. The lane-balanced design, number of lanes on mainline, speed limit, and speed difference present significant heterogeneous effects on crash rates. Findings suggest that the RPMV-Tobit model is a superior approach for comprehensive crash rates modeling and traffic safety evaluation purposes.