Investigation into Human Factors in Rear-End Crashes of Commercial Trucks Based on Human Factors Analysis and Classification System with Bayesian Networks Fusion Method

Abstract

Rear-end crashes of commercial trucks (ReC-CTs) account for the main type of truck traffic crashes, and human factors are important influencing factors that cause ReC-CTs. This study aims to investigate systematic human factors involved in ReC-CTs and further explore relationships between human factors at all levels and induced paths of unsafe acts. In this study, a total of 320 in-depth investigation cases of ReC-CTs in China from 2015 to 2022 were collected, and a novel systematic approach integrating the Human Factors Analysis and Classification System (HFACS) with Bayesian networks (BN) was proposed to identify and quantitatively analyze the human factors of ReC-CTs. The analysis of results leads to the following conclusions: 1) An improved HFACS model was constructed to identify 38 human factors related to ReC-CTs and to conduct a classification analysis at a systemic level; 2) The new systems-based method that integrates HFACS with BN, which can highlight the interrelationships among causal categories at various levels, is an effective method to quantitatively analyze the human factors of ReC-CTs; and 3) The influence relationships between unsafe acts and factors at various levels were quantitatively analyzed at a systemic level; the important influencing factors of each level that lead to unsafe acts were identified, and the most likely induced path for each unsafe act was determined. The research results can provide important guidance for effectively controlling the significant human factors at all levels of HFACS and for the targeted formulation of preventive measures for ReC-CTs.

Keywords

commercial trucks rear-end crashes human factors fusion method Human Factors Analysis and Classification System Bayesian networks

In recent years, with the continuous development of the global economy and the surge in demand for freight transportation, the road freight industry has become increasingly important in the global logistics market ( 1 ). It is undeniable that the increasing busyness of road transportation inevitably leads to road traffic accidents. As reported by the World Health Organization, approximately 135,000 people worldwide die from traffic accidents related to commercial trucks each year. According to statistics from the Traffic Management Science Research Institute of the Ministry of Public Security in China, the death rate of commercial vehicle road traffic accidents per 10,000 vehicles in China was 3.95, which is 2.9 times higher than the average level of other types of vehicle traffic accident, and commercial trucks are the main vehicle type in commercial vehicle accidents. It is reported that rear-end crashes of commercial trucks (ReC-CTs) are the most common type of truck traffic accident, and commercial trucks are mainly of the flat-head type ( 2 ). Once a rear-end crash occurs, the consequences are quite heavy losses. Given the severity of the road safety situation associated with commercial trucks, it is imperative to conduct research on ReC-CTs and to take further preventive measures to avoid similar incidents.

To forestall rear-end crashes and enhance the safety of road traffic, researchers in this domain carried out extensive investigations on the causes of such accidents. Presently, the primary research is based on traffic accident data and statistical analytical methods. By constructing various accident causation models, risk factors of rear-end crashes were analyzed from the perspectives of drivers, vehicles, roads, and the environment. Furthermore, some researchers undertook studies from the perspectives of vehicle driving theory, driver behavior characteristics, and freight company traits. These research results revealed that illegal driving behaviors of drivers (such as speeding, overloading, fatigue driving, and distracted attention), non-standard technical conditions of vehicles, medical conditions, and severe weather are important factors affecting rear-end crashes. It can be seen that the research on how unsafe acts lead to rear-end crashes has been conducted quite comprehensively, and many research results have been achieved. However, there is still relatively little research on how the deeper underlying factors affect unsafe acts. To gain a clearer understanding of the mechanism of rear-end crashes, it is imperative to find a suitable accident causation model and conduct further analysis on the influence of the deeper underlying factors on unsafe acts.

In recent years, numerous scholars have conducted research on system accident causation models, such as the Swiss Cheese Model, the Human Factors Analysis and Classification System (HFACS), the short-term attention memory priority model, and other models ( 3 – 5 ). Compared with other methods, HFACS is presently a mature and practical system that is capable of analyzing and classifying human factors with remarkable results. It not only distinguishes human errors that lead to accidents but also discovers potential risk factors from organizational processes and environments. However, HFACS is only a qualitative analysis model that cannot quantitatively analyze accident risk factors or efficiently identify relationships between factors at various levels. Therefore, it is necessary to integrate the HFACS model with other quantitative research methods to compensate for the limitations of traditional methods in identifying human factors. For example, Chen proposed a mixed HFACS-GRA model to study the important human factors and their causal relationships in maritime accidents ( 6 ). Chen fused HFACS with the Apriori algorithm to determine key human factors in construction industry accidents ( 7 ). Harris and Li combined neural networks with HFACS to predict unsafe acts based on the preconditions of unsafe acts ( 8 ).

The HFACS model is a directed acyclic graph (DAG) with nodes, edges, and weights, similar to a Bayesian network (BN) structure. As a powerful tool for handling uncertain knowledge, BN can perform probabilistic reasoning, identify relationships between variables, and use these relationships to make predictions ( 9 ). Therefore, this study proposes a method that integrates HFACS with BN to investigate and analyze ReC-CTs. The goal is to systematically identify the human factors contributing to rear-end crashes, to quantitatively analyze the interrelationships between these factors, and to provide guidance for accurately formulating preventive measures. The main work of this study is as follows:

1) Revise the original HFACS framework and identify the contributory factors of ReC-CTs using causal categories based on HFACS.

2) Propose a novel systematic approach that integrates HFACS with BN to solve the problem of the inability to quantitatively measure the interrelationships between human factors in ReC-CTs.

3) Quantitatively analyze the causal relationships and intrinsic correlations between human factors in ReC-CTs based on the constructed HFACS-BN model.

Literature Review

To draw on previous research experiences and existing research results both domestically and internationally, and to elevate the study of human factors in ReC-CTs to a higher level, a substantial amount of literature related to human factors in rear-end crashes was reviewed. The following is a summary and overview of relevant research, primarily containing three aspects: human factors in rear-end crashes, HFACS in transportation, and BN applications.

Human Factors in Rear-End Crashes

Scholars both domestically and internationally have conducted extensive research and analysis on the influencing factors of truck rear-end crashes, achieving fruitful results. Xi et al., by establishing a logistic regression model to analyze accident data, investigated the relevant human factors leading to rear-end accidents on highways ( 10 ). The study identified key factors such as driving experience, load, fatigue, driving speed, weather, and accident time affecting rear-end accidents. Wang et al. proposed a random parameter polynomial logit model with heterogeneous mean and variance to examine the differences in human factors when different types of vehicle are involved in rear-end accidents causing varying degrees of damage ( 11 ). Aryan et al. applied both the random parameter binary logit model and support vector machine model to study the main influencing factors of severity in large truck rear-end accidents ( 12 ). Sun et al., through the analysis of 3,982 accident data sets, established a set of factors influencing the severity of rear-end accidents ( 13 ). They used a classification and regression tree model to identify important human factors affecting rear-end accidents. Jung et al., based on negative binomial regression of traffic accidents, quantitatively analyzed the effectiveness of highway rest area additions in reducing rear-end accidents caused by fatigue driving using empirical Bayesian methods ( 14 ). Zhang et al., focusing on fatal truck accidents, used the fault tree method to conclude that a lack of timely vehicle detection and maintenance is a common cause of truck rear-end accidents ( 15 ). Huting et al. utilized random forest and logistic regression models to analyze factors affecting the risk of rear-end accidents, with results indicating a higher likelihood of rear-end accidents when drivers are older, fatigued, inexperienced, and in congested traffic conditions ( 16 ).

The researchers mentioned above have analyzed the influencing factors of rear-end accidents using different mathematical models. However, these studies mainly focus on factors directly related to accidents, such as drivers, vehicles, roads, and the environment, without delving into the impact of freight companies on rear-end accidents. In theory, the management of freight companies may influence drivers, leading to the occurrence of rear-end crashes. Therefore, it is necessary to further investigate the primary influencing factors of freight companies on rear-end crashes.

HFACS in Transportation

Since the introduction of the HFACS model, it has rapidly expanded in the field of aviation safety for investigating and analyzing human factors. Through practical validation, it has proven to be an effective method. Subsequently, the model has gradually extended to other domains, becoming a comprehensive framework with aviation safety at its core. It has been widely applied in various fields such as maritime, railway, coal, and healthcare. In recent years, scholars in the transportation field have conducted numerous studies on human factors based on HFACS, providing beneficial empirical support for the further development of the model.

In the aviation field, Shappell et al. proposed the HFACS model based on the reason theory model, analyzing and summarizing hundreds of U.S. Navy aviation accident reports ( 17 ). They later applied the HFACS model to analyze commercial aviation accidents. The results indicated that HFACS can effectively analyze human errors in the system during accidents. Kilic and Gümüş applied the HFACS framework to analyze 30 nighttime commercial airplane accidents in the past 5 years ( 18 ). They found that the physical environment is the most critical factor, followed by skill errors, and thirdly, perception and decision errors.

In the maritime field, Chauvin et al. revised the HFACS model and analyzed human and organizational factors in 39 maritime ship collision accidents ( 19 ). The research results showed that most ship collision accidents are caused by “decision errors,” and that there is a high frequency of inappropriate planning operations and non-compliance with safety management system regulations by management in accidents. Kim and Seong used the HFACS framework to identify and classify human factors in maritime accidents ( 20 ). The research results demonstrated that the HFACS model can effectively identify potential risk factors in maritime accidents, aiding in formulating safety measures to prevent similar incidents in the future. Chen et al. applied the HFACS framework to systematically analyze human and organizational factors in 3,976 maritime total loss accidents ( 21 ). They selected 11 variables as independent variables to analyze and predict the death probability of maritime total loss accidents.

In the railway field, Madigan et al. used HFACS to analyze 78 reports of minor safety incidents in the UK railways, finding a strong focus on active failures in these reports, with little mention of latent factors ( 22 ). When studying general railway accidents, special attention should be paid to potential factors related to organizational influence and inadequate supervision. Kim and Yoon applied the improved HFACS-Railway Accident (RA) model to analyze 80 railway accident investigation reports in the UK ( 23 ). The results indicated that the HFACS-RA model has strong practicality in analyzing the occurrence process of railway accidents, and that there is a significant correlation between the various components of this model in railway accidents.

In the coal mining field, Patterson and Shappell, to understand human errors and system defects in the mining process, applied HFACS to analyze 508 coal mining accidents in Australia ( 24 ). The results showed that the most common unsafe behavior in coal mining accidents is skill errors. Zhang et al. collected data on 94 major coal mining accidents in China from 1997 to 2011 and conducted a systematic analysis using HFACS ( 25 ). Empirical results showed that the highest frequencies in 5 levels, 14 categories, and 48 indicators are unsafe behavior, inadequate supervision, and inadequate rectification of hidden dangers.

In conclusion, the HFACS model has been widely applied in the analysis of human factors in accidents in various fields such as aviation, maritime, railways, and mining. Its practicality has been validated in these domains. However, the application of the HFACS model in the analysis of human factors in ReC-CTs has not been explored. Therefore, there is an urgent need to use the HFACS model to explore the underlying factors of ReC-CTs and to reasonably categorize the factors.

Bayesian Networks (BN) Applications

BN, as a tool for reasoning and analyzing uncertain knowledge, has demonstrated excellent practicality and ease of operation in studying the mechanisms of human factors in accidents. This method has been widely applied in the field of accident analysis.

Baksh et al. applied the BN model to assess operational risks in navigation in the Arctic region ( 26 ). Combining historical data and expert judgment, they constructed prior probabilities for BN nodes. Through uncertainty and sensitivity analysis, the model determined the likelihood of maritime accidents such as ship collisions, sinking, and grounding. De et al. utilized latent class clustering analysis to preliminarily analyze 3,229 accidents on rural roads in Spain ( 27 ). They identified crucial factors related to the severity of accidents and used BN inference to determine variables most relevant to severe injuries or fatalities. Topuz and Delen proposed a multi-step probability reasoning model based on Bayesian belief networks ( 28 ). Aimed at using a representative accident dataset, the model identified high-risk factors affecting the probability of severe injury in accidents and their apparent significance. Hossain and Muromachi collected road traffic accident data from Japanese highways and employed a random polynomial logit model to identify the most important predictor variables ( 29 ). They applied Bayesian belief networks to establish a real-time collision prediction model. Wang et al. developed a BN-based risk probability assessment model, exploring the probability of accidents under different conditions ( 30 ). They found that driver behavior is the primary factor influencing accidents, followed by vehicle conditions and lighting conditions. Lia et al., based on 2 years of intersection traffic accident data in New Mexico, combined cluster analysis and hierarchical Bayesian models to investigate variables affecting the severity of driver injuries in intersection traffic accidents ( 31 ). Alizadeh et al. established a BN model describing traffic accidents ( 32 ). Through quantitative analysis of accident data, they explored the relationships between accident occurrence and various influencing factors, proposing a novel traffic accident prediction method based on this analysis. Chen et al. constructed a polynomial logic model to study and identify significant factors influencing the severity of rear-end collision driver injuries ( 33 ). They then used identified significant factors to build a BN. The analysis indicated that poor lighting conditions, strong winds, traffic volume, and other factors may significantly increase driver injuries in rear-end crashes.

In summary, the BN model addresses the limitations in explaining the dependency relationships among accident factors and can achieve quantitative analysis of accident risk factors and prediction of accident consequences. However, current research is mainly based on accident investigation report data, combining domain expert knowledge to construct BN structures. The conditional probability tables (CPTs) for each node are then obtained through mathematical and statistical methods, leading to lower accuracy in the BN model. Secondly, the current BN model primarily focuses on analyzing human factors involved in the accident occurrence process, namely the factors of unsafe acts and preconditions for unsafe acts in the HFACS framework. Its main purpose is to analyze how unsafe acts lead to accidents and predict the probability of accidents. However, there is limited in-depth research at the systemic level on the influence of enterprise-management-related factors on unsafe acts and their interdependencies, urging the need for further studies to address the existing research gaps.

Overall, previous research on the influencing factors of ReC-CTs mainly focuses on the analysis of factors such as drivers, vehicles, roads, and environment. This overlooks the potential impact of corporate management on ReC-CTs and fails to deeply analyze the relationship between drivers’ unsafe behavior and corporate management. This study introduces the HFACS model for the first time in the analysis of ReC-CTs and enhances its applicability. Through a systematic analysis of corporate management and drivers’ personal factors, we identify the key factors influencing ReC-CTs caused by corporate management. Additionally, BN analysis is employed to quantitatively analyze the interrelationships among influencing factors, providing reference for preventing the occurrence of ReC-CTs and further reducing the risks and losses associated with such accidents.

Materials and Methods

It is necessary to have an overview of human factors in ReC-CTs based on the HFACS integrated BN method. Firstly, we collected in-depth investigation reports on ReC-CTs to provide a data foundation for subsequent research. Next, we combined grounded theory with the HFACS framework to identify the human factors leading to ReC-CTs. Subsequently, based on the identified human factors, we extracted the necessary data from accident reports. Finally, we constructed the HFACS-BN model to quantitatively analyze the interaction relationships between human factors in ReC-CTs. The process integrates established methodologies to systematically analyze and understand the underlying human factors contributing to ReC-CTs.

Data Sources

High-quality accident data is key to ensuring the reliability of research conclusions. In this study, the accident data were obtained from the traffic police department of Shandong Province and the national vehicle accident in-depth investigation organization in China. The accident samples were strictly screened according to the following criteria: 1) rear-end crashes involving two parties only; 2) accident samples including various types of commercial trucks, such as light trucks (n = 89), medium trucks (n = 122), heavy trucks (n = 227), and semi-trailers (n = 202), excluding micro trucks; and 3) accident case data with complete information to ensure that the research analysis conditions are met. Based on the selection criteria, 320 in-depth investigation reports of ReC-CTs that occurred between 2015 and 2022 were selected as analysis samples. These in-depth investigation reports were compiled by accident investigation teams comprising members from the local safety supervision bureau, public security bureau, traffic police detachment, transportation bureau, and in-depth investigation organization, after conducting thorough investigations into the accidents. These reports provide detailed descriptions of the information of vehicles and personnel related to the accidents, the consequences of the accidents, the analysis of direct and indirect causes, the deficiencies of enterprise management, the determination of accident responsibility, and the prevention suggestions. These data possess high reliability and applicability, contributing to ensuring the accuracy and credibility of research conclusions.

Human Factors Analysis and Classification System (HFACS)

HFACS was proposed based on the Swiss Cheese Model (SCM), which is a widely acknowledged framework for classifying human errors ( 34 ). The SCM model addresses human errors and their underlying factors at four levels: unsafe acts (active failures), preconditions for unsafe acts, unsafe supervision, and organizational influences (latent failures). Figuratively speaking, these failures are like “holes” at different levels of the “cheese” and, when the holes at different levels line up in a straight line, danger can pass through all the holes in the defense measures and lead to accidents. However, the SCM model is only a theoretical framework and does not provide the specific tools needed for accident investigation. To bridge the gap between theory and practice, the HFACS proposes a formal and structurally consistent framework with the SCM, providing detailed information on active and latent failures at all levels, helping accident investigators and analysts systematically identify organizational deficiencies, as shown in Figure 1 ( 35 ).

The original HFACS framework was primarily used by the U.S. military to investigate and analyze human factors in aviation accidents. To enhance the applicability of the original HFACS framework for analyzing ReC-CTs, this study revises the various levels of causal factors. At the same time, to mitigate the subjectivity associated with expertise in identifying human factors, this study introduces grounded theory to refine the human factors. Grounded theory is a qualitative research method that is spiral and bottom-up ( 36 ). Based on the collected data and materials, a complete set of theories is finally summarized through bottom-up screening and refinement. Therefore, grounded theory is not a method for obtaining data, but a method for in-depth analysis and organization of data. The coding process of grounded theory consists of three stages: open coding, axial coding, and selective coding ( 37 ). Firstly, the concepts and categories are defined through open coding. Secondly, these categories are clustered into primary categories through axial coding. Finally, the relationship between the primary categories is clarified through selective coding, culminating in the identification of the core categories. The application of grounded theory to facilitate the identification of unsafe acts in accidents based on the HFACS framework has been previously demonstrated. Therefore, it is scientifically feasible to use grounded theory to identify systematic human factors in ReC-CTs.

Bayesian Networks (BN)

Bayesian networks (BN) were originally proposed by Pearl and have become one of the most effective theoretical models in the fields of uncertain knowledge representation and theoretical reasoning ( 38 ). Compared with other quantitative accident analysis methods such as the analytic hierarchy process, event tree, and fuzzy comprehensive evaluation, BN offers the distinctive advantage of probabilistic inference ( 38 ). The BN model represents causal relationships and probability distributions in the objective world in a unique way, and is structurally represented as a DAG that comprises of node variables and directed edges connecting these nodes. Each node represents a variable state, while the directed edges represent dependencies between variables. The strength of correlation or confidence coefficient between variables is expressed through the CPT.

The construction of the BN model comprises two distinct processes: BN structure learning and BN parameter learning. The purpose of BN structure learning is to determine the factor variables (nodes) relevant to the research object and the dependence or independence relationships between each node, which is the basis of BN learning. The purpose of BN parameter learning is to determine the CPT of each node, to quantify the dependence relationship between the child node and the parent node. Figure 2 is a simple BN model, where node “A” is a parent node, and nodes “B” and “C” are child nodes. Based on the data-driven approach, this study uses the data training module of Netica software to train the model by importing accident data, and obtains the CPT of each node.

However, BN has seldom been utilized to analysis human factors of traffic accidents, despite the wide application in the traffic accident prediction. Therefore, this study aims to use BN as a suitable method to quantitatively analyze the relationships of human factors in ReC-CTs, and to explore the induced paths of unsafe acts.

Data Coding

To meet the needs of data analysis, it is necessary to extract the required data from the in-depth accident investigation reports. Firstly, for the sake of convenient documentation, the accident investigation reports can be numbered according to the type of vehicles involved and the time of the accident, such as “01-light truck VS semi-trailer-20190518.” Then, the accident reports can be decomposed and encoded in accordance with the revised HFACS model, and the required coding data for commercial truck accidents is obtained. For the purposes of this study, the coding for each factor is limited to “Y” or “N,” with the coding protocol being as follows: if there is any element in the accident report case that meets the definition of a certain element in the HFACS model, we will record that element as “Y,” otherwise, we will record it as “N.”

Considering that the decomposition and coding results of the in-depth accident investigation report directly affect the effectiveness and reliability of the subsequent data analysis, this study develops a comprehensive and detailed coding scheme. Firstly, to enhance the consistency and reliability of coding, the coding guidelines were drafted before coding, ensuring that all coders can understand and adhere to the same coding rules. Additionally, to eliminate subjective judgments by coders during the coding process based on their knowledge and experience, the coding work was jointly completed by three coders with sufficient work experience and appropriate academic backgrounds in the field of traffic accidents: a senior professor (C1), a professional accident investigator (C2), and a traffic accident judicial appraisal worker (C3). They received formal training before coding to ensure their comprehension and conformity with the coding rules. Secondly, to ensure the effectiveness and consistency of the data coding by the three coders, it is necessary to perform a consistency check of the coded data after the completion of coding by each coder individually. Considering that the coding data is a binary variable and the number of coders is three, the Kappa test method was used to check the consistency of the coding data. As the Kappa test can only be performed on the coded data of two coders at a time, this study necessitates three consistency tests. Finally, although the coded data consistency test results of the three coders were reliable, their coded data were not necessarily identical. Therefore, the final coded values for accident causes will be determined based on the coded data recognized by at least two coders.

Combination of HFACS and BN

To clearly describe how HFACS and BN can be effectively integrated, a systematic research framework has been proposed, as shown in Figure 3.

Step 1. The structural model of the BN should be established based on the HFACS framework to visualize the correlations between human factors. In this study, the causal relationships between adjacent hierarchical risk factors within the HFACS framework were determined by using SPSS software to calculate χ², P, odds ratio (OR) values, and 95% confidence intervals. Subsequently, these relationships were transformed into a Bayesian network structure.

Step 2. Determining the probability tables for each node. Based on the encoded data, this study utilizes the data training module of Netica software to automatically calculate the CPTs for each node. Furthermore, it identifies the predictive rules of the BN model.

Step 3. Examining the validity of the HFACS-BN model. This study employs randomly selected sampled data to validate the model. By inputting the data of A-, B-, and C-level nodes of the validation sample into the model, the prediction values of the D-level nodes are observed to determine whether they match the true values of the validation sample, and the degree of consistency is used to represent the effectiveness of the model.

The HFACS-BN model established through the above steps can be used to diagnose and analyze the human factors in ReC-CTs.

Results

Identification of Human Factors Based on HFACS

In this study, 320 in-depth accident investigation reports were analyzed within the framework of HFACS using the three-level coding method of grounded theory. A total of 38 human factors leading to ReC-CTs were identified and categorized into 13 categories and 4 levels. The data encoding work has been completed and a consistency test was conducted to ensure the accuracy of the coding data—the results are shown in Table 1. Everitt revealed the Cohen’s Kappa coefficient is considered moderate between 0.40 and 0.60, and satisfactory above 0.60 ( 39 ). The results show that the Kappa coefficient of consistency between any two coders’ coding data is between 0.826 and 0.928, indicating a high level of consistency. To better illustrate the various nodes of the HFACS model, Table 2 provides detailed explanations for each human factor.

Table 1.

Identified Human Factors of Rear-End Crashes of Commercial Trucks with Human Factors Analysis and Classification System Framework

Factors and codes	Factors name	Kappa Coefficient			Number	Percentage
Factors and codes	Factors name	C1 and C2	C2 and C3	C3 and C1	Number	Percentage
Organizational influences (A)
Resource management
A1	Inadequate investment in safety management	0.914	0.924	0.935	142	44.38
A2	Improper management of driver qualification	0.856	0.872	0.859	54	16.88
A3	Improper management of transportation equipment	0.831	0.826	0.845	77	24.06
Organizational climate
A4	Insufficient safety culture climate	0.915	0.898	0.908	135	42.19
A5	Unfair corporate policies	0.804	0.852	0.886	54	16.88
A6	Inadequate risk control policies	0.837	0.864	0.829	72	22.50
Organizational process
A7	Insufficient implementation of safety management	0.941	0.934	0.945	130	40.63
A8	Imperfect safety rules and regulations	0.831	0.853	0.875	110	34.38
A9	Unclear responsibility for safety management	0.847	0.836	0.856	78	24.38
A10	Loopholes in safety supervision system	0.891	0.882	0.879	106	33.13
Unsafe supervision (B)
Inadequate supervision
B1	Lack of regulatory policies and institutions	0.826	0.853	0.834	114	35.63
B2	Lack of attention to supervision and management	0.942	0.938	0.944	136	42.50
B3	Insufficient training and education	0.925	0.940	0.943	162	50.63
Planned inappropriate operations
B4	Unreasonable operation tasks	0.852	0.896	0.878	65	20.31
B5	Unreasonable driver scheduling	0.855	0.828	0.847	33	10.31
Failed to correct known problems
B6	Failed to report unsafe tendencies	0.843	0.851	0.864	85	26.56
B7	Failure to trace the hidden danger rectification	0.836	0.851	0.846	47	14.69
Supervisory violations
B8	Intentional violation of rules and regulations	0.882	0.876	0.863	56	17.50
B9	Failed to implement rules and regulations	0.885	0.894	0.899	90	28.13
Preconditions for unsafe acts (C)
Driver factors
C1	Poor physical and mental state	0.914	0.919	0.932	92	28.75
C2	Insufficient driving experience and skills	0.875	0.869	0.884	42	13.13
C3	Poor safety awareness	0.931	0.919	0.925	178	55.63
C4	Obstructing safe driving behaviors	0.908	0.924	0.918	114	35.63
C5	Insufficient qualifications	0.862	0.858	0.849	28	8.75
Physical environmental factors
C6	Poor operating environment	0.899	0.915	0.907	87	27.19
C7	Poor vehicle technical conditions	0.833	0.845	0.867	93	29.06
Unsafe acts (D)
Skill-based errors
D1	Improper operation	0.842	0.837	0.826	52	16.25
D2	Incorrect use of light signals	0.884	0.897	0.876	28	8.75
Decision errors
D3	Wrong response to emergency	0.828	0.831	0.839	33	10.31
D4	Improper emergency measures	0.883	0.865	0.852	40	12.50
Perceptual errors
D5	Visual illusion	0.851	0.865	0.878	48	15.00
D6	Unsafe driving distance	0.938	0.934	0.941	166	51.88
D7	Distracted attention	0.942	0.925	0.934	108	33.75
Violations
D8	Fatigue driving	0.933	0.938	0.929	136	42.50
D9	Speeding	0.911	0.907	0.910	115	35.94
D10	Overloading	0.938	0.936	0.936	145	45.31
D11	Parking violations	0.921	0.934	0.927	68	21.25
D12	Low-speed driving	0.898	0.905	0.910	72	22.50

Table 2.

Human Factors Description

Factor	Description
Organizational influences
Inadequate investment in safety management	Transportation enterprises do not invest sufficient funds in safety management to maintain and enhance the safety of the system.
Improper management of driver qualification	Transportation enterprises have issues with the certification, training, and supervision of drivers.
Improper management of transportation equipment	There are issues in enterprises concerning the certification, training, and supervision of drivers.
Insufficient safety culture climate	Transportation enterprises lack emphasis and investment in safety culture, resulting in an overall weak safety management atmosphere.
Unfair corporate policies	The salary, resource allocation, and vehicle scheduling policies formulated by transportation enterprises are unfair, leading to drivers facing unjust conditions.
Inadequate risk-control policies	Transportation enterprises have insufficient or imperfect policies, regulations, or measures in place to address various risks.
Insufficient implementation of safety management	Transportation enterprises have inadequate or imperfect execution of safety management measures, leading to ineffective safety management outcomes.
Imperfect safety rules and regulations	The regulations and systems pertaining to safety management in enterprises are incomplete, lacking specificity, or containing loopholes, and therefore fail to effectively guide and regulate safety management work.
Unclear responsibility for safety management	The responsibility delineation for safety management among management personnel in transportation enterprises is unclear or ambiguous.
Loopholes in safety supervision system	The deficiencies or shortcomings in the execution of safety supervision duties by regulatory authorities lead to insufficient, ineffective, or lenient safety management.
Unsafe supervision
Lack of regulatory policies and institutions	The lack of comprehensive policies, regulations, and regulatory systems by regulatory authorities results in inadequate, disordered, or ineffective supervision work.
Lack of attention to supervision and management	Transportation enterprises lack sufficient attention and investment in supervisory management, leading to inadequate supervision, ineffective regulatory measures, and unsatisfactory regulatory outcomes.
Insufficient training and education	Transportation enterprises provide insufficient, inadequate, or untimely safety training and education.
Unreasonable operation tasks	Transportation enterprises may assign transportation tasks to drivers in an unreasonable or unscientific manner.
Unreasonable driver scheduling	Transportation enterprises may have unreasonable or unscientific scheduling, shift rotation, or work arrangements for drivers.
Failure to report unsafe tendencies	Supervisory personnel fail to promptly report to management or relevant departments the presence of safety hazards, unsafe behavior, or tendencies toward unsafe conditions.
Failure to trace the hidden danger rectification	After identifying safety hazards, transportation enterprises fail to promptly and effectively rectify, investigate, and supervise them.
Intentional violation of rules and regulations	Personnel deliberately and intentionally fail to comply with, or violate, the internal management systems, regulations, and rules of transportation enterprises.
Failure to implement rules and regulations	Transportation enterprises fail to effectively implement or enforce these rules and regulations in practical operations.
Preconditions for unsafe acts
Poor physical and mental state	Driver’s physical health or mental state problems lead to decreased work efficiency, lack of concentration, decreased work quality, and other situations.
Insufficient driving experience and skills	Insufficient experience and skill level of drivers in driving vehicles.
Poor safety awareness	Drivers have insufficient understanding of driving safety issues.
Obstructing safe driving behaviors	A series of unsafe behaviors affecting driver safety.
Insufficient qualifications	The driver lacks the necessary qualifications, abilities, or skills.
Poor operating environment	Unfavorable environmental conditions encountered during vehicle operation can adversely affect driving safety.
Poor vehicle technical conditions	The vehicle itself has technical issues or defects.
Unsafe acts
Improper operation	Drivers did not operate the vehicle correctly according to traffic regulations and safe driving principles.
Incorrect use of light signals	Drivers did not use the vehicle’s lighting system correctly according to traffic regulations during driving.
Wrong response to emergency	Wrong judgments or inappropriate decisions made by drivers in the face of unexpected or emergency situations.
Improper emergency measures	When facing emergencies or urgent situations, the measures taken by drivers are inappropriate or inadequate.
Visual illusion	Misinterpretations or misunderstandings that occur during the driver’s visual perception process can affect their judgment of road conditions and the traffic environment.
Unsafe driving distance	Driver did not maintain a sufficient safe distance from the preceding vehicle during the driving process.
Distracted attention	The driver’s attention is drawn to multiple different things or tasks, resulting in an inability to concentrate fully on a single task or objective.
Fatigue driving	After long periods of continuous driving, drivers experience a physiological and psychological imbalance, leading to an objective decline in driving skills.
Speeding	The behavior of drivers driving on the road at a speed exceeding the prescribed speed limit.
Overloading	Vehicles are driven when they exceed their design or statutory load limits.
Parking violations	Parking vehicles in prohibited or restricted parking areas constitutes a violation of traffic management regulations.
Low-speed driving	Vehicles traveling at low speeds on the road, typically well below the speed limit or the normal flow of traffic.

Based on the coding data, the frequency of each human factor was listed in Table 1. The total frequency of human factors in the 320 ReC-CTs is 3,391. On average, 11 factors have been identified in each accident. Among the four levels of HFACS, the highest frequency is attributed to “unsafe acts (n = 1,011),” accounting for 29.81% of the total frequency (n = 3,391), followed by “organizational influences (n = 958),”“unsafe supervision (n = 788),” and “preconditions for unsafe acts (n = 634),” which account for 23.25%, 23.24%, and 18.7% of the total frequency, respectively. Among the 13 categories in HFACS, “violations (n = 536),”“driver factors (n = 454),”“organizational process (n = 424),” and “inadequate supervision (n = 412)” were significantly more frequent than the other categories. Within the 38 human factors considered, “poor safety awareness (n = 178),”“unsafe driving distance (n = 166),” and “insufficient training and education (n = 162)” appeared with more significant frequency.

Determination of HFACS-BN Model

The chi-square test and OR analysis was conducted between adjacent levels of human factors in HFACS framework to determine the causal relationships between human factors. Bender et al. indicated that if there is a relationship with $P < 0.05$ and $OR > 1$ between any two factors in adjacent levels, it is considered that there is a correlation between the two factors ( 40 ). These two factors can be regarded as two nodes connected by a directed edge in the BN, with the directed edge pointing from the upper-level factor to the lower-level associated factor. Based on the correlation analysis of human factors, the results with P < 0.05 and OR > 1 were selected, as shown in Table 3. The statistical results indicate that there are 46 pairs of associated factors in adjacent levels of HFACS, with the association strength index (OR) ranging from 4.224 to 21.856, indicating a strong correlation.

Table 3.

The Factors Correlations Analysis of Adjacent Levels

Related factors of adjacent level	$χ^{2}$ test results		Odds ratio	95% confidence interval
Related factors of adjacent level	$χ^{2}$	$P$	Odds ratio	95% confidence interval
A×B
A1×B3	5.012	0.000	3.522	1.649–4.315
A1×B7	7.421	0.005	6.898	1.532–28.684
A1×B9	15.279	0.002	3.018	1.763–6.524
A4×B2	5.381	0.008	6.651	3.283–13.461
A4×B7	7.548	0.010	6.957	1.542–28.356
A4×B9	6.483	0.006	15.412	4.584–46.852
A7×B2	5.354	0.021	2.037	1.252–3.084
A7×B3	4.682	0.031	5.863	1.465–30.251
A7×B4	10.629	0.001	4.982	1.353–18.306
A10×B4	5.127	0.007	6.320	1.324–52.364
A10×B7	4.625	0.001	8.642	1.028–12.355
A10×B9	8.861	0.018	7.859	1.476–15.476
B×C
B2×C1	6.038	0.015	8.652	1.504–23.683
B2×C2	5.204	0.019	5.261	1.086–13.522
B2×C3	6.659	0.009	3.052	1.493–10.957
B3×C2	9.208	0.006	5.332	1.375–13.574
B3×C3	6.584	0.012	6.522	1.358–16.852
B3×C4	10.341	0.002	3.689	1.350–10.283
B4×C1	4.552	0.014	4.536	1.166–18.654
B4×C6	3.962	0.017	3.854	1.968–14.096
B7×C1	6.225	0.016	2.206	1.053–7.543
B7×C6	4.254	0.028	3.582	1.115–10.256
B9×C1	6.682	0.015	6.255	1.374–18.868
B9×C3	8.869	0.008	8.066	2.156–24.385
C×D
C1×D5	5.824	0.015	4.224	2.205–15.624
C1×D7	8.365	0.006	14.252	2.368–28.659
C1×D8	10.322	0.008	9.568	3.084–30.526
C1×D11	5.085	0.020	4.854	1.086–17.898
C2×D1	8.656	0.005	7.858	1.574–22.266
C2×D4	5.897	0.018	7.528	1.549–30.082
C2×D6	7.656	0.026	9.586	3.542–20.056
C2×D9	8.975	0.004	6.356	1.852–18.058
C3×D6	10.533	0.002	16.628	3.055–28.545
C3×D8	4.550	0.032	4.544	1.086–16.627
C3×D9	8.806	0.004	6.508	2.426–15.233
C3×D10	10.406	0.001	14.667	2.983–33.729
C3×D11	5.132	0.026	7.556	1. 405–16.588
C4×D11	16.887	0.001	21.856	4.508–38.652
C4×D1	9.136	0.005	7.856	2.265–16.878
C4×D5	12.525	0.025	10.368	7.538–32.025
C4×D6	10.408	0.003	14.665	2.858–32.552
C4×D7	13.661	0.000	14.989	4.282–36.125
C6×D5	5.184	0.032	4.765	1.285–18.068
C6×D6	10.533	0.001	16.554	3.025–30.854
C6×D9	17.101	0.000	21.052	3.858–38.561
C6×D11	9.203	0.028	6.485	1.762–18.525

According to the correlations between human factors in adjacent levels obtained from HFACS, a visualized model of the HFACS-BN network structure was constructed using the Netica software. The model includes 4 levels, 23 nodes, and 46 directed edges. After determining the BN structure, the data training module of Netica software was used to establish the CPT of each node by importing parameter learning training set (288 cases of in-depth investigation coding data). The trained BN is shown in Figure 4.

A random selection of 32 in-depth investigation coding data of ReC-CTs was used for model validation, and the validation results are presented in Table 4. It can be concluded that the BN model constructed has a high degree of coincidence under the current accident sample size conditions, and it is feasible to apply this model for quantitative analysis of human factors in ReC-CTs.

Table 4.

The Model Validation Results

Coincidence (P)	P = 100%	80% < P < 100%	P < 80%
Number of samples	16	14	2

Quantitative Analysis of Human Factors Based on BN Inference

Single-Factor Diagnostic Analysis Results

Diagnostic (abductive reasoning) analysis refers to obtaining the cause of the result and the posterior probability of the nodes given some new evidence ( 41 ). The fault diagnosis function of the BN allows the analysis of potential influencing factors leading to various unsafe acts. By setting the probability values of various factors at D-level and running the BN, we record the changes in probability values of the corresponding factors at the upper levels (A, B, C level). The influence degree of the upper-level factors on the D-level factors is determined based on the magnitude of the changes in probability value. For example, by setting the probability value of the state “Y” of “D6” to 100%, the changes in probability values of upper-level factors can be obtained, as shown in Figure 5. We repeat above process for the other factors in D-level and finally obtain the changes in probability value for factors at levels A, B, and C. After organizing the data, we obtain Table 5.

Table 5.

The Probability Change Value for Each Node

D-level factors	A-level factors				B-level factors					C-level factors
D-level factors	A7	A4	A1	A10	B2	B3	B4	B7	B9	C2	C4	C1	C6	C3
D1	1.6	0.7	3.2	0.2	0.3	4.8	0.2	0.1	0.2	6.2	4.3	1.6	1.3	0.8
D4	1.2	0.2	4.3	0.4	0.5	5.0	0.1	0.2	0.3	5.7	2.6	1.3	2.2	0.6
D5	0.8	0.8	0.9	2.9	1.4	0.6	2.5	0.6	1.1	2.2	7.4	5.3	4.6	1.1
D6	4.9	2.0	3.5	0.8	5.2	6.2	1.3	0.4	0.6	0.7	3.2	3.1	1.5	7.9
D7	1.5	4.2	1.1	0.5	1.2	0.6	2.8	1.1	1.6	0.8	5.5	8.6	2.1	1.3
D8	5.3	3.6	1.4	2.2	1.3	0.7	2.1	2.2	2.3	0.4	0.2	6.5	1.0	8.2
D9	3.1	1.5	2.8	1.0	0.9	2.6	1.7	2.0	0.4	2.3	0.5	0.6	3.8	9.4
D10	2.9	1.4	1.2	2.6	2.2	1.5	2.6	1.6	1.4	0.2	0.0	0.1	0.0	6.4
D11	0.9	0.5	2.5	0.3	0.7	1.8	1.4	0.5	0.6	1.2	0.5	2.2	3.5	7.0

To intuitively reflect the influence degree of factors in the upper levels (A, B, and C) on the factors in D-level. The curves were plotted based on the data in Table 5, as shown in Figure 6. Figure 6a shows that: A7 (insufficient implementation of safety management) has the most significant impact on D6 (unsafe driving distance) and D8 (fatigue driving); D7 (distracted attention) is greatly affected by A4 (insufficient safety culture climate); A1 (inadequate investment in safety management) has a significant impact on D4 (improper emergency measures), D1 (improper operation), D9 (speeding), and D11 (parking violations); and D5 (visual illusion) is heavily influenced by A10 (loopholes in safety supervision system). Figure 6b shows that: B2 (lack of attention to supervision and management) has the greatest impact on D6; B3 (insufficient training and education) affects D1, D4 (improper emergency measures), and D6; B4 (unreasonable operation tasks) has a significant impact on D5, D7, and D10; D8, D9, and D10 are greatly affected by B7 (failure to trace the hidden danger rectification); and B9 (failed to implement rules and regulations) also has a significant impact on D8 and D10. Figure 6c shows that: C2 (insufficient driving experience and skills) has a significant impact on D1; C4 (obstructing safe driving behaviors) deeply affects D5; C1 (poor physical and mental state) has a significant impact on D5 and D7; D5, D9, and D11 are greatly affected by C6 (poor operating environment); and C3 (poor safety awareness) has a greatest significant impact on D6, D8, D9, D10, and D11.

All-Factors Diagnostic Analysis Results

After analyzing the influence of the upper-levels factors on the various factors in the D level, it is necessary to further analyze their impact on the entire level of unsafe acts. To achieve this, the probability of all factors in the D level is set to 100%, and, running the BN model, the results are shown in Figure 7. Based on the analysis results in Figure 7, the distributions of the influence of each upper-level factor on the entire D level are shown in Figure 8. Figure 8 shows that the upper-level factors with the greatest influence on the entire D level are A7 (insufficient implementation of safety management), B3 (insufficient training and education), C3 (poor safety awareness), C1 (poor physical and mental state), and C4 (obstructing safe driving behaviors). Furthermore, it can be noted that the influence of the upper levels (A, B, and C) on D level gradually increases, indicating that the closer to the unsafe acts, the greater the influence.

The Induced Paths of Unsafe Acts

The induced paths of unsafe acts refer to potential routes or process that trigger the occurrence of unsafe acts ( 42 ). By analyzing the induced paths of unsafe acts, the organization can identify potential factors and processes, and take measures to reduce the incidence of accidents as early as possible. According to the diagnostic inference results of unsafe acts factors (Table 5 and Figure 6), the potential factors with the greatest probability value changes in each upper-level factor are sequentially sought, and the most likely induced paths for unsafe acts are obtained, as shown in Table 6.

Table 6.

The Induced Paths of Unsafe Acts

Unsafe act	Induced path
D1	A1→B3→C2 (C4) →D1
D4	A1→B3→C2→D4
D5	A10→B4→C4 (C1, C6) →D5
D6	A7 (A1) →B3 (B2) →C3→D6
D7	A4→B4→C1 (C4) →D7
D8	A7 (A4) →B9 (B7, B4) →C3 (C1) →D8
D9	A7 (A1) →B3→C3→D9
D10	A7 (A10) →B4 (B2) →C3→D10
D11	A1→B3→C3→D11
All factors of D level	A7→B3→C3 (C1, C4) →unsafe acts

Discussion

Unsafe acts are the direct cause of ReC-CTs and represent the primary focus of current research ( 43 ). The findings of this study indicate that “unsafe driving distance,”“overloading,” and “fatigue driving” are significant unsafe acts leading to ReC-CTs, all of which constitute illegal acts. McDonald et al. discovered that driver violations are the primary cause of road traffic accidents ( 44 ). As a result, government departments should strengthen their inspection and control efforts during key periods and road sections, which can timely detect and prevent illegal behaviors such as overloading and fatigue driving. At the same time, big data and intelligent recognition technologies should be used to monitor the status of drivers and eliminate the phenomenon of fatigue driving. Furthermore, it is recommended to conduct traffic safety propaganda and education from multiple angles, enhancing the traffic safety awareness of transportation companies, drivers, and their social groups (family, friends, etc.). These measures will help minimize the occurrence of unsafe acts, thereby preventing ReC-CTs from occurring.

Merely studying unsafe acts is far from sufficient. It is imperative to delve into the underlying causes of such acts. The comprehensive diagnostic analysis reveals that the preconditions for unsafe acts have the most significant impact on unsafe acts. Among these acts, “poor safety awareness” is the most critical factor, followed by “obstructing safe driving acts” and “poor physical and mental state.” Earl found that behavior is determined by consciousness; traffic safety consciousness influences and governs a driver’s driving behaviors ( 45 ). A weak traffic safety consciousness is bound to give rise to traffic violations, making it essential to enhance the traffic safety awareness of drivers to prevent accidents. At the same time, the number of traffic violations and accidents caused by obstructing safe driving behaviors has been increasing year by year, with the use of mobile phones, smoking, and eating being the most common behaviors ( 46 ). To realize real-time monitoring and early warning of unsafe driving behavior, the installation of a real-time driver status monitoring system for commercial trucks is essential. Additionally, it is necessary to strengthen the health monitoring of drivers, eliminate the phenomenon of driving while ill, and reasonably arrange transportation tasks to ensure adequate rest, which is a crucial measure to prevent ReC-CTs.

From a regulatory perspective, “insufficient training and education” is the most significant influencing factor. At present, most of the freight vehicles in China are operated by affiliated companies, which lack sufficient attention to driver safety training and education ( 47 ). At the same time, the drivers of commercial trucks generally have a low level of education and limited understanding of safety knowledge. As a result, some drivers have weak traffic safety awareness and insufficient driving skills. Therefore, road transport enterprises must strengthen and standardize traffic safety promotion and education as a long-term task. To address safety training and education, a scientifically accurate training plan and content should be developed to encourage drivers to comprehensive master traffic safety regulations and driving skills, while a sound training assessment mechanism should be established to ensure the mastery of safety knowledge and the implementation of the training effectiveness. The aforementioned measures can comprehensively improve drivers’ safety knowledge, driving skills, and safety awareness, thereby reducing or avoiding the occurrence of unsafe acts and accidents.

At the highest level of “organizational influences” of HFACS, “insufficient implementation of safety management,”“inadequate investment in safety management,” and “insufficient safety culture climate” are important influencing factors. At present, to save operating costs, transport enterprises are devoid of safety management personnel and investment. The safety management system is virtually non-existent, and the management work is lax or even blank. As a result, some vehicles operate illegally and drivers work with illness. The safety culture climate is extremely poor ( 48 ). Therefore, transport enterprises should establish a sound regulatory system, implement regulatory responsibilities, strengthen safety supervision and management, track vehicle performance and driver qualifications, and realize dynamic supervision of drivers and transport vehicles. At the same time, the government and transportation management departments should also strengthen the supervision of road transport enterprises, promptly handle and punish illegal and irregular acts, fundamentally improve the safety management situation of freight enterprises, and create a good safety culture atmosphere.

The results of this study confirm that there are significant relationships between the upper-level factors and the lower-level factors of HFACS. To avoid ReC-CTs with minimal cost at an earlier stage, the important influencing factors at each level and the interrelationships between the levels should be fully understood, and measures should be taken as far as possible at the higher levels of HFACS. Additionally, Table 6 lists the induced paths of unsafe acts, which are important relationships between the causal factors of ReC-CTs. We can formulate prevention strategies according to the probability of each human factor and the correlation degree of the induced paths, which can provide a precise “road map” for preventing such accidents. Lv et al. found that accountability measures have a positive impact on improving the induced paths of unsafe acts, while organizational influence indicators ultimately affect drivers’ unsafe acts ( 49 ). Establishing a sound accountability system can prompt individuals and organizations to recognize the consequences of their actions and strengthen their sense of responsibility, thereby reducing the occurrence of unsafe behaviors. Meanwhile, the implementation of an accountability system can swiftly identify and rectify unsafe acts, safety hazards, and accidents in daily work, contributing to the prevention of accidents. Therefore, organizations should establish correct values, take effective accountability measures, strengthen internal supervision and management of the enterprise, improve the safety awareness of enterprise managers and drivers, and cut off the most likely induced path as soon as possible, thereby reducing or even eliminating the occurrence of ReC-CTs.

Conclusions

The core work of this study is to develop a new fusion method that integrates the advantages of HFACS and BN, which can comprehensively and objectively analyze human factors and their related relationships in ReC-CTs. The main conclusions are as follows:

1) The HFACS framework was revised with the support of grounded theory, and 13 categories and 38 human factors were identified from four levels: organizational influences, unsafe supervision, preconditions for unsafe acts, and unsafe acts. Thereby, an HFACS model suitable for the analysis of human factors in ReC-CTs was constructed.

2) The correlations between factors at adjacent levels of the HFACS model were obtained through the SPSS analysis approach. A comprehensive approach that integrates HFACS with BN has been proposed. The influence relationships between unsafe acts and each factor at upper levels were quantitatively analyzed from system level based on the HFACS-BN model, and the most likely induced paths for unsafe acts were obtained. In the later stages, ReC-CTs can be prevented by cutting off the critical links and paying special attention to the key influencing factors.

3) This study found that factors at the level of unsafe acts accounted for the largest proportion of all human factors and should be the key control level of ReC-CTs. At the same time, “driver factors,”“inadequate supervision,” and “organizational processes” were the key factors that led to unsafe acts, and recommendations were made to intervene in the upper-level factors of HFACS and to strengthen the control of the key factors. Additionally, this study also demonstrated that the influence of upper-level factors on the level of unsafe acts increases gradually, and that effective control of upper-level factors of HFACS and taking measures to eliminate potential accidents at an earlier stage are conducive to nipping hazards in the bud.

This work also has its limitations. Firstly, concerning the construction of the HFACS framework, since the original HFACS was designed for analyzing aviation accident causes, there may be some subjectivity involved in modifying its indicators for the context of ReC-CTs, which may weaken its universality. Secondly, the relatively small sample sizes of ReC-CTs may affect the accuracy of some estimations, and increasing the sample sizes in the future can improve the precision of the model. Overall, this study provides meaningful insights and important support for formulating more accurate prevention countermeasures for ReC-CTs.

Footnotes

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: Du, X.J. and W. Zhao; data collection: Zhao, W; analysis and interpretation of results: Du, X.J; draft manuscript preparation: Du, X.J. and W. Zhao. All authors reviewed the results and approved the final version of the manuscript.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: the Fundamental Research Funds for the Central Universities (No. 2572023AW66), the Key Research and Development Guiding Projects of Heilongjiang Province (No. GZ20220027), and the National Science Fund for Young Scholars (No. 51108068).

References

Kumar

Anbanandam

Assessment of Environmental and Social Sustainability Performance of the Freight Transportation Industry: An Index-Based Approach. Transport Policy, Vol. 124, 2022, pp. 43–60.

Zhang

Y. C.

Liu

Cao

H. N.

Zhang

Wang

G. H.

Optimization of Aerodynamic Drag Coefficient for a Cabover Truck Considering the Effects of Crosswind. Automotive Engineering, Vol. 43, No. 5, 2021, pp. 713–720.

Reinach

Viale

Application of Human Error Framework to Conduct Train Accident/Incident Investigations. Accident Analysis and Prevention, Vol. 38, No. 2, 2006, pp. 396–406.

Ergai

Cohen

Sharp

Wiegmann

Gramopadhye

Assessment of the Human Factors Analysis and Classification System (HFACS): Intra-Rater and Inter-Rater Reliability. Safety Science, Vol. 82, 2016, pp. 393–398.

Revell

K. M. A.

Allison

Sears

Stanton

N. A.

Modelling Distributed Crewing in Commercial Aircraft with STAMP for a Rapid Decompression Hazard. Ergonomics, Vol. 62, No. 2, 2019, pp. 156–170.

Chen

S. T.

An Approach of Identifying the Common Human and Organisational Factors (HOFs) Among a Group of Marine Accidents Using GRA and HFACS-MA. Journal of Transportation Safety & Security, Vol. 12, No. 10, 2019, pp. 1252–1294.

Chen

The Development and Validation of a Human Factors Analysis and Classification System for the Construction Industry. International Journal of Occupational Safety and Ergonomics, Vol. 28, No. 1, 2022, pp. 479–493.

Harris

W. C.

Using Neural Networks to Predict HFACS Unsafe Acts from the Pre-Conditions of Unsafe Acts. Ergonomics, Vol. 62, No. 2, 2019, pp. 181–191.

D’Urso

Vitale

Bayesian Networks Model Averaging for Bes Indicators. Social Indicators Research, Vol. 151, No. 3, 2020, pp. 897–919.

10.

J. F.

Guo

H. Y.

Tian

Liu

L. S.

Sun

W. F.

Analysis of Influencing Factors for Rear-End Collision on the Freeway. Advances in Mechanical Engineering, Vol. 11, No. 7, 2019, pp. 1–10.

11.

Wang

C. Z.

Chen

Zhang

Y. L.

Cheng

J. C.

Analysis of Injury Severity in Rear-End Crashes on an Expressway Involving Different Types of Vehicles Using Random-Parameters Logit Models with Heterogeneity in Means and Variances. Transportation Letters-the International Journal of Transportation Research, Vol. 15, No. 7, 2022, pp. 742–753.

12.

Aryan

Amin

Ali

Investigating Factors Affecting Severity of Large Truck-Involved Crashes: Comparison of the SVM and Random Parameter Logit Model. Journal of Safety Research, Vol. 77, 2021, pp. 151–160.

13.

Sun

Z. Y.

Wang

J. Y.

Chen

Y. Y.

H. P.

Influence Factors on Injury Severity of Traffic Accidents and Differences in Urban Functional Zones: The Empirical Analysis of Beijing. International Journal of Environmental Research and Public Health, Vol. 15, No. 12, 2019, p. 2722.

14.

Jung

Joo

Evaluating the Effects of Supplemental Rest Areas on Freeway Crashes Caused by Drowsy Driving. Accident Analysis & Prevention, Vol. 99, 2017, pp. 356–363.

15.

Zhang

Kecojevic

Komljenovic

Investigation of Haul Truck-Related Fatal Accidents in Surface Mining Using Fault Tree Analysis. Safety Science, Vol. 65, No. 6, 2014, pp. 106–117.

16.

Huting

Reid

Nwoke

Identifying Factors That Increase Bus Accident Risk by Using Random Forests and Trip-Level Data. Transportation Research Record: Journal of the Transportation Research Board, 2016. 2539: 149–158.

17.

Shappell

Detwiler

Holcomb

Hackworth

Boquet

Wiegmann

D. A.

Human Error and Commercial Aviation Accidents: An Analysis Using the Human Factors Analysis and Classification System. Human Factors, Vol. 49, No. 2, 2007, pp. 227–242.

18.

Kilic

Gümüş

Application of HFACS to the Nighttime Aviation Accidents and Incidents. Journal of Aviation, Vol. 4, No. 2, 2020, pp. 10–16.

19.

Chauvin

Lardjane

Morel

Clostermann

J. P.

Langard

Human and Organisational Factors in Maritime Accidents: Analysis of Collisions at Sea Using the HFACS. Accident Analysis & Prevention, Vol. 59, 2013, pp. 26–37.

20.

Kim

H. T.

Seong

Development of a Human Factors Investigation and Analysis Model for Use in Maritime Accidents: A Case Study of Collision Accident Investigation. Journal of Navigation and Port Research, Vol. 41, No. 5, 2017, pp. 303–318.

21.

Chen

Bian

Wan

Wang

Zheng

Cheng

Factor Assessment of Marine Casualties Caused by Total Loss. International Journal of Disaster Risk Reduction, Vol. 47, 2020, p.101560.

22.

Madigan

Golightly

Madders

Application of Human Factors Analysis and Classification System (HFACS) to UK Rail Safety of the Line Incidents. Accident Analysis & Prevention, Vol. 97, 2016, pp. 122–131.

23.

Kim

D. S.

Yoon

W. C.

An Accident Causation Model for the Railway Industry: Application of the Model to 80 Rail Accident Investigation Reports From the UK. Safety Science, Vol. 60, 2013, pp. 57–68.

24.

Patterson

J. M.

Shappell

S. A.

Operator Error and System Deficiencies: Analysis of 508 Mining Incidents and Accidents from Queensland, Australia Using HFACS. Accident Analysis & Prevention, Vol. 42, No. 4, 2010, pp. 1379–1385.

25.

Zhang

Jing

Bai

A Systems Approach to Extraordinarily Major Coal Mine Accidents in China From 1997 to 2011: An Application of the HFACS Approach. International Journal of Occupational Safety and Ergonomics, Vol. 25, No. 2, 2019, pp. 181–193.

26.

Baksh

A. A.

Abbassi

Garaniya

Marine Transportation Risk Assessment Using Bayesian Network: Application to Arctic Waters. Ocean Engineering, Vol. 159, 2018, pp. 422–436.

27.

O. J.

López

Mujalli

Analysis of Traffic Accidents on Rural Highways Using Latent Class Clustering and Bayesian Networks. Accident Analysis & Prevention, Vol. 51, 2013, pp. 1–10.

28.

Topuz

A Probabilistic Bayesian Inference Model to Investigate Injury Severity in Automobile Crashes. Decision Support Systems, Vol. 150, 2021, pp. 687–694.

29.

Hossain

Muromachi

A Bayesian Network Based Framework for Real Time Crash Prediction on the Basic Freeway Segments of Urban Expressways. Accident Analysis & Prevention, Vol. 45, 2012, pp. 373–381.

30.

Wang

Zhu

Bayesian Network-Based Risk Assessment for Hazmat Transportation on the Middle Route of the South-to-North Water Transfer Project in China. Stochastic Environmental Research and Risk Assessment, Vol. 30, No. 3, 2016, pp. 841–857.

31.

Lia

Z. N.

Chen

Y. S.

Examining Driver Injury Severity in Intersection-Related Crashes Using Cluster Analysis and Hierarchical Bayesian Models. Accident Analysis & Prevention, Vol. 120, 2018, pp. 139–151.

32.

Alizadeh

S. S.

Mortazavi

S. B.

Sepehri

M. M.

Prediction of Vehicle Traffic Accidents Using Bayesian Networks. Pure and Applied Sciences, Vol. 3, No. 6, 2014, pp. 356–362.

33.

Chen

Zhang

G. H.

Tarefder

A Multinomial Logit Model-Bayesian Network Hybrid Approach for Driver Injury Severity Analyses in Rear-End Crashes. Accident Analysis & Prevention, Vol. 80, 2015, pp. 76–88.

34.

Shappell

S. A.

Wiegmann

D. A.

A Human Error Approach to Aviation Accident Analysis: The Human Factors Analysis and Classiﬁcation System. Ashgate, England, 2012.

35.

Cohen

Wiegmann

Shappell

Evaluating the Reliability of the Human Factors Analysis and Classification System. Aerospace Medicine and Human Performance, Vol. 86, No. 8, 2015, pp. 728–735.

36.

Nelson

L. K.

Computational Grounded Theory: A Methodological Framework. Sociological Methods & Research, Vol. 49, No.1, 2020, pp. 3–42.

37.

Turner

Astin

Grounded Theory: What Makes a Grounded Theory Study. European Journal of Cardiovascular Nursing, Vol. 20, No. 3, 2021, pp. 285–289.

38.

Geiger

Pearl

Logical and Algorithmic Properties of Independence and Their Application to Bayesian Networks. Annals of Mathematics and Artificial Intelligence, Vol. 2, No. 1–4, 1990, pp. 165–178.

39.

Everitt

Making Sense of Statistics in Psychology: A Second-Level Course. Oxford University Press, Oxford, UK, 1996.

40.

Bender

Grouven

Using Binary Logistic Regression Models for Ordinal Data with Non-Proportional Odds. Journal of Clinical Epidemiology, Vol. 51, No. 10, 2019, pp. 809–816.

41.

Kabir

Papadopoulos

Applications of Bayesian Networks and Petri Nets in Safety, Reliability, and Risk Assessments: A Review. Safety Science, Vol. 115, 2019, pp. 154–175.

42.

Ghasemi

Kalatpour

Moghimbeigi

Mohhamadfam

A Path Analysis Model for Explaining Unsafe Behavior in Workplaces: The Effect of Perceived Work Pressure. International Journal of Occupational Safety and Ergonomics, Vol. 24, No. 2, 2018, pp. 303–310.

43.

Lattanzi

Freschi

Machine Learning Techniques to Identify Unsafe Driving Behavior by Means of In-Vehicle Sensor Data. Expert Systems with Applications, Vol. 176, 2021, p. 114818.

44.

McDonald

Berecki-Gisolf

Stephan

Newstead

Traffic Offending and Deterrence: An Examination of Recidivism Amongst Drivers in Victoria, Australia Born Prior to 1975. PLoS One, Vol. 15, No. 10, 2020, p. e0239942.

45.

Earl

The biological function of consciousness. Frontiers in Psychology, Vol. 5, 2014, p. 697.

46.

Waddell

L. P.

Wiener

K. K. K.

What’s Driving Illegal Mobile Phone Use? Psychosocial Influences on Drivers’ Intentions to Use Hand-Held Mobile Phones. Transportation Research Part F: Traffic Psychology and Behaviour, Vol. 22, 2014, pp. 1–11.

47.

Cantor

D. E.

Corsi

T. M.

Grimm

C. M.

Ozpolat

A Driver Focused Truck Crash Prediction Model. Transportation Research Part E: Logistics and Transportation Review, Vol. 46, No. 5, 2010, pp. 683–692.

48.

Nævestad

T. O.

Elvebakk

Phillips

R. O.

The Safety Ladder: Developing an Evidence-Based Safety Management Strategy for Small Road Transport Companies. Transport Reviews, Vol. 38, No. 3, 2018, pp. 372–393.

49.

Zhao

Dai

D. S.

The Construction-Agent-System Project Governance Effect in the Context of Chinese New Infrastructure Construction. Journal of Shandong University (Philosophy and Social Sciences), No. 1, 2021, pp. 175–184.