Transferability of accident prediction models

Accident prediction models, the vast majority of which are negative binomial regression models, are of considerable importance to highway agencies since they can be used to conduct many traffic safety studies. However, not every agency possesses sufficient accident statistics that enable it to develop reliable models of its own. This problem gives rise to interest in the transferability of accident prediction models in time and space. It would save time, effort, and money if accident prediction models developed for one region in one period of time could be applied in different time periods and regions to produce reliable safety studies.This paper presents methods for recalibrating negative binomial accident models before transferring them for use in different time periods and regions of space. The paper emphasizes that the recalibration of the shape parameter of a transferred model using local data is absolutely necessary. It explains that it is also desirable to recalibrate the constant term of the transferred model in order to allow the model to better suit local conditions. A moment method is presented for recalibrating the shape parameter of a transferred model when its constant term is not recalibrated. However, a maximum likelihood method is presented for recalibrating both the shape parameter and the constant term of the transferred model and is shown to be superior to the recalibration methods existing in the traffic safety literature.     

Random parameter models for accident prediction on two-lane undivided highways in India

Introduction

Generalized linear modeling (GLM), with the assumption of Poisson or negative binomial error structure, has been widely employed in road accident modeling. A number of explanatory variables related to traffic, road geometry, and environment that contribute to accident occurrence have been identified and accident prediction models have been proposed. The accident prediction models reported in literature largely employ the fixed parameter modeling approach, where the magnitude of influence of an explanatory variable is considered to be fixed for any observation in the population. Similar models have been proposed for Indian highways too, which include additional variables representing traffic composition. The mixed traffic on Indian highways comes with a lot of variability within, ranging from difference in vehicle types to variability in driver behavior. This could result in variability in the effect of explanatory variables on accidents across locations. Random parameter models, which can capture some of such variability, are expected to be more appropriate for the Indian situation.

Method

The present study is an attempt to employ random parameter modeling for accident prediction on two-lane undivided rural highways in India. Three years of accident history, from nearly 200 km of highway segments, is used to calibrate and validate the models.

Results

The results of the analysis suggest that the model coefficients for traffic volume, proportion of cars, motorized two-wheelers and trucks in traffic, and driveway density and horizontal and vertical curvatures are randomly distributed across locations.

Conclusions

The paper is concluded with a discussion on modeling results and the limitations of the present study.     

Road casualties and enforcement: distributional assumptions of serially correlated count data

OBJECTIVE: Road safety data are often in the form of counts and usually temporally correlated. The objective of this research is to investigate the distributional assumptions of road safety data in the presence of temporal correlation. METHODS: Using the generalized linear model framework, four distributional assumptions are considered: normal, Poisson, quasi-Poisson and negative binomial, and appropriate models are estimated. Monthly casualty and police enforcement data from Greece for a period of six years (January 1998-December 2003) have been used. The developed models include sinusoidal latent terms to capture the temporal serial correlation of observations. Several statistical goodness-of-fit diagnostic tests have been performed for the results of the estimated models, and the predictive capabilities of the models are investigated. RESULTS: The residuals of the quasi-Poisson and negative binomial models do not show any serial correlation. The signs of the estimated coefficients for all models are consistent and intuitive. In particular, a negative coefficient value for the number of breath alcohol controls indicates that the number of persons killed and seriously injured decreases as the intensity of breath alcohol controls increases. The Poisson model fails to capture the overdispersion in the data, thus underestimating the standard errors of the estimated coefficients. CONCLUSION: The results suggest that the quasi-Poisson and negative binomial outperform the normal and Poisson models in this application. The findings of this research demonstrate a clear link between the intensification of police enforcement and the reduction of traffic accident casualties. In particular, an increase in the number of breath alcohol controls in Greece after 1998 contributed to a reduction in the number of persons killed and seriously injured from traffic accidents.     

Investigating the effects of monthly weather variations on Connecticut freeway crashes from 2011 to 2015

IntroductionThe objective of this research is to investigate the effects of monthly weather conditions on traffic crash experience on freeways, considering the interactions between weather, traffic volumes, and roadway conditions. Methods: Data from the state of Connecticut from 2011to 2015 were used. Random parameters negative binomial models with first-order, autoregressive covariance were estimated for representative types of freeway crashes (front-to-rear, sideswipe-same-direction, and fixed-object), most severe crashes (i.e., fatal and injury crashes), and non-injury crashes (i.e., property-damage-only crashes). Results: Major findings are that variations in monthly traffic volumes, roadway geometry, and weather conditions explain much of the variations in monthly traffic crashes. Time effects exist in the panel monthly data for all types of crashes. Taking into account this effect improves model prediction results. When the raw weather measures are highly correlated, using dimension reduction techniques helps to extract more interpretable weather factors. By considering the interaction effects between roadway condition variables, additional findings were found. In general, lower temperature, more heavy fog days, decreased precipitation, lower wind speed, higher monthly traffic volumes, and narrower inside shoulder were found to be associated with higher monthly crashes. The effects of area type and outside shoulder width change dramatically as the number of through lanes changes. Practical applications: The findings of this research could help researchers and general readers gain a better understanding of the effects of monthly weather conditions and other roadway factors on freeway crashes and give engineers practical guidelines on improving freeway safety.     

Exploring relationships between microscopic kinetic parameters of tires under normal driving conditions,road characteristics and accident types

IntroductionFreeway accidents are a leading cause of death in China, which also triggers substantial economic loss and an emotional burden to society. However, the internal mechanism of how microscopic kinetic parameters of vehicles influenced by road characteristics determine the occurrence of different types of accidents has not been explicitly studied. This research aimed to explore the “link role” of tire microscopic kinetic parameters in road characteristic variables and traffic accidents to aid in facilitating the traffic design and management, and thus to prevent traffic accident. Method: A mountain freeway in Zhejiang Province, China was used as the research object and the data used in this paper were obtained through a real-time vehicle experiment. Multiple estimation models, including the standard ordered logit (SOL) model, fixed parameters logit (FPL) model, and random parameters logit (RPL) model were established. Results: The findings show that road characteristics will affect the longitudinal kinetic characteristics of the vehicle and, consequently, map the level of risk of rear-end accidents. Driving compensation effects were also identified in this paper (i.e., the drivers tend to be more cautious in complicated driving circumstances). Another finding relating to the mountain freeway is that different tunnel characteristics (e.g., tunnel entrance and tunnel exit) have different effects on different types of traffic accidents. Practical Applications: The framework proposed in this article can provide new insight for researchers to enlarge the research subjects of both explanatory and outcome variables in accident analysis. Future research could be implemented to consider more driving conditions.     

An algorithm for assessing the risk of traffic accident

INTRODUCTION: This study is aimed at developing an algorithm to estimate the number of traffic accidents and assess the risk of traffic accidents in a study area. METHOD: The algorithm involves a combination of mapping technique (Geographical Information System (GIS) techniques) and statistical methods (cluster analysis and regression analysis). Geographical Information System is used to locate accidents on a digital map and realize their distribution. Cluster analysis is used to group the homogeneous data together. Regression analysis is performed to realize the relation between the number of accident events and the potential causal factors. Negative binomial regression model is found to be an appropriate mathematical form to mimic this relation. Accident risk of the area, derived from historical accident records and causal factors, is also determined in the algorithm. The risk is computed using the Empirical Bayes (EB) approach. A case study of Hong Kong is presented to illustrate the effectiveness of the proposed algorithm. RESULTS: The results show that the algorithm improves accident risk estimation when comparing to the estimated risk based on only the historical accident records. The algorithm is found to be more efficient, especially in the case of fatality and pedestrian-related accident analysis. IMPACT ON INDUSTRY: The output of the proposed algorithm can help authorities effectively identify areas with high accident risk. In addition, it can serve as a reference for town planners considering road safety.     

道路交通事故发展趋势分析与预测

根据福建省2000 -2010年交通事故相关指标,采用统计图表分析法进行交通事故发展趋势分析与安全水平比较研究,结果表明交通事故各项绝对指标总体呈下降趋势,但从万车死亡率、受伤人数与死亡人数比及交通事故死亡人数占各类事故死亡人数比重等相对指标看,交通安全总体水平偏低,交通事故后果比较严重.对交通事故死亡人数与GDP、机动车保有量、公路通车里程、人口数四项影响因素进行了多元线性回归分析,分析得出四个影响因素总体对交通事故死亡人数的线性影响是显著的,采取向后筛选策略线性回归分析得出,死亡人数与GDP的线性关系是显著的,根据回归结果建立了交通事故的预测模型.     

Application of generalized link functions in developing accident prediction models

The Bayesian Poisson–Gamma hierarchy, leading to the negative binomial distribution, has been the standard practice in developing accident prediction models. To linearize the relationship connecting the mean of the negative binomial distribution to relevant covariates, a canonical log link has traditionally been used. Typically, little information is available regarding the choice of a particular link. To avoid link misspecification, it is proposed to nest the canonical log link model within a generalized link family and subsequently use the full Bayes method for parameter estimation, performance evaluation and inference. The proposed approach was applied to a sample of accident and traffic volume data corresponding to 99 intersections in the city of Edmonton, Alberta. The results showed that both the generalized link model and the traditional canonical link model provided adequate fit to the data. However, the Bayes factor provided a clear statistical support for the use of the generalized link approach. A procedure for link validation is also described. It allows the users (e.g., road authorities) to consider the changes in predicted accidents that will result if a generalized link is used instead of a canonical link. If a certain maximal change is tolerated, the canonical link can be used to analyze the data; otherwise the generalized link is worth the extra efforts and should be adopted. When compared with the traditional approach, the generalized link model was found to predict a lower number of accidents whenever there is a heavy traffic at the major approach, especially if combined with light flow on the minor approach. The paper concludes by identifying out areas for further research.     

Assessing the likelihood of secondary crashes on freeways with Adaptive Signal Control System deployed on alternate routes

Introduction: Reducing the likelihood of freeway secondary crashes will provide significant safety, operational and environmental benefits. This paper presents a method for assessing the likelihood of freeway secondary crashes with Adaptive Signal Control Systems (ASCS) deployed on alternate routes that are typically used by diverted freeway traffic to avoid any delay or congestion due to a freeway primary crash. Method: The method includes four steps: (1) identification of secondary crashes, (2) verification of alternate routes, (3) assessment of the likelihood of secondary crashes for freeways with ASCS deployed on alternate routes and non-ASCS (i.e. pre-timed, semi- or fully-actuated) alternate routes, and (4) investigation of unobserved heterogeneity of the likelihood of freeway secondary crashes. Four freeway sections (i.e., two with ASCS deployed on alternate routes and two non-ASCS alternate routes) in South Carolina are considered. Results and Conclusions: Findings from the logistic regression modeling reveal significant reduction in the likelihood of secondary crashes for one freeway section (i.e., Charleston I-26 E) with ASCS deployed on alternate route. Other factors such as rear-end crash, dark or limited light, peak period, and annual average daily traffic contribute to the likelihood of freeway secondary crashes. Furthermore, random-parameter logistic regression model results for Charleston I-26 E reveal that unobserved heterogeneity of ASCS effect exists across the observations and ASCS are associated with the reduction of the likelihood of freeway secondary crashes for 84% of the observations (i.e., primary crashes). Location of the primary crash on the freeway is observed to affect the benefit of ASCS toward freeway secondary crash reduction as the primary crash’s location determines how many upstream freeway vehicles will be able to take the alternate route. Practical Applications: Based on the findings, it is recommended that the South Carolina Department of Transportation (SCDOT) considers deploying ASCS on alternate routes parallel to freeway sections where high percentages of secondary crashes are found.     

高速公路交通应急救援资源的配置

针对交通事故与救援资源存在随机性的特点,建立了交通救援资源配置的随机模型。根据高速公路特有的道路条件与交通机理,确定了模型中救援服务水平、事故概率的权重,配置点至事故点的权值以及随机资源等参数值。将所建立的模型与参数用于河南省高速公路的救援资源配置,研究表明,新的配置方案较好满足了交通事故高概率区域的救援需求,并且合理减少了救援需求区域的资源配置数量,为目前高速公路交通应急资源的科学配置提供了重要的参考依据。     

Applying machine learning approaches to analyze the vulnerable road-users' crashes at statewide traffic analysis zones

Introduction: In this paper, we present machine learning techniques to analyze pedestrian and bicycle crash by developing macro-level crash prediction models. Methods: We collected the 2010–2012 Statewide Traffic Analysis Zone (STAZ) level crash data and developed rigorous machine learning approach (i.e., decision tree regression (DTR) models) for both pedestrian and bicycle crash counts. To our knowledge, this is the first application of DTR models in the burgeoning macro-level traffic safety literature. Results: The DTR models uncovered the most significant predictor variables for both response variables (pedestrian and bicycle crash counts) in terms of three broad categories: traffic, roadway, and socio-demographic characteristics. Additionally, spatial predictor variables of neighboring STAZs were considered along with the targeted STAZ in both DTR models. The DTR model considering spatial predictor variables (spatial DTR model) were compared without considering spatial predictor variables (aspatial DTR model) and the model comparison results discovered that the prediction accuracy of the spatial DTR model performed better than the aspatial DTR model. Finally, the current research effort contributed to the safety literature by applying some ensemble techniques (i.e. bagging, random forest, and gradient boosting) in order to improve the prediction accuracy of the DTR models (weak learner) for macro-level crash count. The study revealed that all the ensemble techniques performed slightly better than the DTR model and the gradient boosting technique outperformed other competing ensemble techniques in macro-level crash prediction models.     

基于模糊均生函数和最优子集回归的飞行事故率预测

为有效预防飞行事故的发生,针对飞行事故率具有随机波动性和趋势性的特点,采用模糊均生函数(FMGF)和最优子集回归(OSR)建立飞行事故率预测模型。该方法把FMGF延拓序列作为预测因子加入OSR方程,将FMGF分析和因子筛选相结合作OSR,进而对飞行事故率进行预测。通过对美国空军1988—2004年的飞行事故率进行拟合预测,结果表明:将FMGF模型和OSR模型有机结合,能够有效刻画飞行事故率的随机波动特性,并且其预测结果的相对误差也较小。     

基于模糊Petri网的交通事故致因建模分析

针对目前我国道路交通事故多发的现状,以模糊Petri网为工具,以对道路交通安全有重大影响的关键因素为基础,模拟给出其因果关系,建立了基于模糊Petri网的道路交通事故致因分析模型,设计最长路径算法分析模型中某个致因要素导致交通事故发生的可信度。最长路径上的致因要素即为最易引起交通事故的主要因素。案例分析表明,这个模型能够体现各因素之间的逻辑关系,达到了通过数量指标分析道路交通事故主要原因的目的。     

云南省交通事故数据的分析研究

通过对云南省1981—2003年的交通事故统计数据的分析研究,给出了交通事故死亡人数的预测模型。通过与发达国家类似的交通事故历史数据的对比分析,给出以时间和机动车拥有量为自变量、交通事故死亡人数为因变量的简单预测模型,该模型对2004年的交通事故死亡人数的预测是准确的;同时采用该模型预测了云南省交通事故死亡人数的峰值及其年份。结论指出:基于目前的人、车、路和管理水平及发展趋势,云南省的交通事故死亡人数在2013—2018年之间将达到高峰,高峰时的交通事故死亡人数在5528～7369人之间。     

Highway-rail grade crossings accident prediction using Zero Inflated Negative Binomial and Empirical Bayes method

Introduction: Recently the Federal Railroad Administration (FRA) released a new model for accident prediction at railroad grade crossings using a Zero Inflated Negative Binomial (ZINB) model with Empirical Bayes (EB) adjustments for accident history (2). This new model is adopted from the work that was conducted by the authors (3–6). The unique feature of the new FRA model is that it has a single equation for all three warning devices (crossbuck, flashing light, and gates) and uses the same variables regardless of the warning devices at the crossing. Since the New FRA model incorporates the warning device category as one of the variables in its model equation, the predicted accident frequency is higher when a crossing has crossbucks than flashing lights, and higher when it has flashing lights than gates. While this model is significantly better than the old USDOT model (7), its shortcoming is that the single equation does not accurately represent the field condition. Method: This paper presents the ZINEBS model (Zero Inflated Negative binomial with Empirical Bayes adjustment System). The ZINEBS model gives three different equations depending on the type of warning device used at the crossings (gates, flashing lights, and crossbucks). The three equations use variables, some of which are common across all warning devices, while other variables are specific to a warning device. The predicted values for the ZINEBS model show a closer agreement with the field data than the new FRA model. This observation was true for all three warning device types analyzed. Practical Applications: Based on the results of this study, the ZINEBS compliments the new FRA model and should be used when the single equation is not adequately representing the role of traffic control device types and relevant variables associated with that device type.     

On-Scene Injury Severity Prediction (OSISP) Algorithm for Truck Occupants

Objective: The aim of this study is to develop an on-scene injury severity prediction (OSISP) algorithm for truck occupants using only accident characteristics that are feasible to assess at the scene of the accident. The purpose of developing this algorithm is to use it as a basis for a field triage tool used in traffic accidents involving trucks. In addition, the model can be valuable for recognizing important factors for improving triage protocols used in Sweden and possibly in other countries with similar traffic environments and prehospital procedures.

Methods: The scope is adult truck occupants involved in traffic accidents on Swedish public roads registered in the Swedish Traffic Accident Data Acquisition (STRADA) database for calendar years 2003 to 2013. STRADA contains information reported by the police and medical data on injured road users treated at emergency hospitals. Using data from STRADA, 2 OSISP multivariate logistic regression models for deriving the probability of severe injury (defined here as having an Injury Severity Score [ISS] > 15) were implemented for light and heavy trucks; that is, trucks with weight up to 3,500 kg and ??16,500 kg, respectively. A 10-fold cross-validation procedure was used to estimate the performance of the OSISP algorithm in terms of the area under the receiver operating characteristic curve (AUC).

Results: The rate of belt use was low, especially for heavy truck occupants. The OSISP models developed for light and heavy trucks achieved cross-validation AUC of 0.81 and 0.74, respectively. The AUC values obtained when the models were evaluated on all data without cross-validation were 0.87 for both light and heavy trucks. The difference in the AUC values with and without use of cross-validation indicates overfitting of the model, which may be a consequence of relatively small data sets. Belt use stands out as the most valuable predictor in both types of trucks; accident type and age are important predictors for light trucks.

Conclusions: The OSISP models achieve good discriminating capability for light truck occupants and a reasonable performance for heavy truck occupants. The prediction accuracy may be increased by acquiring more data. Belt use was the strongest predictor of severe injury for both light and heavy truck occupants. There is a need for behavior-based safety programs and/or other means to encourage truck occupants to always wear a seat belt.     

Investigating hazardous factors affecting freeway crash injury severity incorporating real-time weather data: Using a Bayesian multinomial logit model with conditional autoregressive priors

Introduction: It has been demonstrated that weather conditions have significant impacts on freeway safety. However, when employing an econometric model to examine freeway crash injury severity, most of the existing studies tend to categorize several different adverse weather conditions such as rainy, snowy, and windy conditions into one category, “adverse weather,” which might lead to a large amount of information loss and estimation bias. Hence, to overcome this issue, real-time weather data, the value of meteorological elements when crashes occurred, are incorporated into the dataset for freeway crash injury analysis in this study. Methods: Due to the possible existence of spatial correlations in freeway crash injury data, this study presents a new method, the spatial multinomial logit (SMNL) model, to consider the spatial effects in the framework of the multinomial logit (MNL) model. In the SMNL model, the Gaussian conditional autoregressive (CAR) prior is adopted to capture the spatial correlation. In this study, the model results of the SMNL model are compared with the model results of the traditional multinomial logit (MNL) model. In addition, Bayesian inference is adopted to estimate the parameters of these two models. Result: The result of the SMNL model shows the significance of the spatial terms, which demonstrates the existence of spatial correlation. In addition, the SMNL model has a better model fitting ability than the MNL model. Through the parameter estimate results, risk factors such as vertical grade, visibility, emergency medical services (EMS) response time, and vehicle type have significant effects on freeway injury severity. Practical Application: According to the results, corresponding countermeasures for freeway roadway design, traffic management, and vehicle design are proposed to improve freeway safety. For example, steep slopes should be avoided if possible, and in-lane rumble strips should be recommended for steep down-slope segments. Besides, traffic volume proportion of large vehicles should be limited when the wind speed exceeds a certain grade.     

高速公路隧道群交通事故风险致因分析

通过对高速公路隧道群特殊地理环境所造成的交通事故后果风险与救援时间的相关分析,提出隧道群交通事故风险致因模型,该模型将隧道群交通事故风险划分3个风险阶段,即初始事故风险、事故发展风险、最终事故风险,不同阶段的风险值受人的因素、车辆因素、隧道群环境因素和防灾救援能力大小的影响而改变。通过对隧道群的风险因素分析,认为隧道群在两毗邻隧道间将可能产生烟雾风险,驾驶人员视觉快速转换的照明风险,以及交通事故防灾控制风险;同时统计的隧道群防灾救援时间概率分布表明,救援队伍能在一定时间内快速到达事故现场,并对较晚到达事故现场救援情况,基于风险分析而提出相应的防范对策和措施。     

高速公路收费站安全评价模型

在研究高速公路收费站交通事故及其安全影响因素相互作用的基础上,选择收费广场渐变率、收费站服务水平和收费广场入口段纵坡作为模型参数,以收费站年交通事故发生次数为模型的标度,通过样本数据的回归分析建立高速公路收费站安全评价模型。经实地采集的收费站事故数据验证表明:该模型用于高速公路主线收费站交通事故预测具有良好的效果;可以评价收费站安全水平并有针对性地提出改善方案,从而降低收费公路事故率。     

Factors associated with driving under the influence of alcohol

Objective: This study examined the risk factors of driving under the influence of alcohol (DUI) among drivers of specific vehicle categories (DSC). On the basis of this research, the variables related to DUI and involvement in traffic crashes were defined. The analysis was conducted for car drivers, bicyclists, motorcyclists, bus drivers, and truck drivers.

Method: The research sample included drivers involved in traffic crashes on the territory of Serbia in 2016 (60,666). Two types of analyses were conducted in this study. Logistic regression established the correlation between DUI and DSC and the The Technique for Order of Preference by Similarity to Ideal Solution (Multi-criteria decision making) method was applied to consider the scoring and explore the potential for the prevalence of DUI on the basis of 2 data sets (DUI and non DUI).

Results: The study results showed that driver error and male drivers were the 2 most significant risk factors for DUI, with the highest scores and potential for prevalence. The nonuse of restraint systems, driver experience, and driver age are the factors with a significant prediction of involvement in an accident and an insignificant prediction of DUI.

Conclusions: Following the development of the logistic prediction models for DUI drivers, testing of the model was conducted for 3 control driver groups: Car, motorcycle, and bicycle. The prediction model with a probability greater than 50% showed that 77% of car drivers were under the influence of alcohol. Similarly, the prediction percentage for motorcyclists and bicyclists amounted to 71 and 67%, respectively. The recommendation of the study is that drivers whose DUI probability is above 50% should be potentially suspected of DUI. The results of this study can help to understand the problem of DUI among specific driver categories and detect DUI drivers, with the aim of creating successful traffic safety policy.