Systematic Review of Quantitative Risk Quantification Methods in Construction Accidents

Abstract

Construction accidents pose significant risks to workers and the public, affecting industry productivity and reputation. While several reviews have discussed risk assessment methods, recent advancements in artificial intelligence (AI), big data analytics, and real-time decision support systems have created a need for an updated synthesis of the quantitative methodologies applied in construction safety. This study systematically reviews the literature from the past decade, following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. A thorough search identified studies utilizing statistical analysis, mathematical modeling, simulation, and artificial intelligence (AI). These methods were categorized and analyzed based on their effectiveness and limitations. Statistical approaches, such as correlation analysis, examined relationships between variables, while mathematical models, like factor analysis, quantified risk factors. Simulation methods, such as Monte Carlo simulations, explored risk dynamics and AI techniques, including machine learning, enhanced predictive modeling, and decision making in construction safety. This review highlighted the strengths of handling large datasets and improving accuracy, but also noted challenges like data quality and methodological limitations. Future research directions are suggested to address these gaps. This study contributes to construction safety management by offering an overview of best practices and opportunities for advancing quantitative risk assessment methodologies.

1. Introduction

Construction is one of the most hazardous industries in the world, accounting for a large proportion of occupational fatalities and injuries [1,2]. Despite the efforts to improve construction safety performance, the accident rate in construction remains high compared to other industries [2,3]. Therefore, it is essential to identify and assess the factors that influence the accident risk level in construction projects and implement effective measures to prevent or mitigate the potential hazards [4,5,6,7].

Traditionally, construction projects employed qualitative risk assessment methods, relying on expert judgment and experience to identify and evaluate risks [8]. While valuable, these methods often lack objectivity and may overlook underlying quantitative factors contributing to accident occurrence. Quantitative approaches, on the other hand, offer a more objective and data-driven perspective by utilizing statistical analysis, mathematical modeling, simulation, and artificial intelligence [9,10,11]. Recent advancements in AI, such as machine learning and deep learning, have shown promise in enhancing predictive accuracy and providing insights into complex risk factors [12]. These methods aim to measure, predict, and optimize the accident risk level based on various data sources, such as accident records, safety inspections, workers’ behaviors, and environmental conditions. Quantitative methods can provide objective and reliable results that can support decision making and risk management in construction safety.

However, the application and development of quantitative methods for accident risk level analysis in construction face several challenges, such as data availability, quality, and validity; methodological limitations and assumptions; practical implications; and the generalization of the results [13]. While several studies have investigated the methods and data sources used for accident risk level analysis in construction, a systematic review of the literature is necessary to provide a comprehensive understanding of the current state of knowledge. Moreover, there is a lack of a comprehensive and systematic review of the existing literature on quantitative methods for accident risk level analysis in construction, which hinders the understanding of the current state of the art, the identification of the research gaps, and the direction for future research. This review adopts the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines to ensure a rigorous and transparent approach to the systematic review process [14]. PRISMA provides a structured framework that enhances the quality and reproducibility of systematic reviews, making it an essential tool in synthesizing research findings [14].

Therefore, the main research question of this review is: What are the methods, data sources, and results of quantitative accident risk level analysis in construction? The objectives of this review are to:

• Conduct a systematic review of the literature on quantitative methods for accident risk level analysis in construction, following the PRISMA guidelines.
• Compare and contrast the different methods, data sources, and results of quantitative accident risk level analysis in construction across the selected studies.
• Evaluate the strengths and limitations of the current literature and suggest directions for future research.

This review contributes to the existing knowledge and practice of construction safety management by:

• Providing a comprehensive and up-to-date overview of the quantitative methods for accident risk level analysis in construction.
• Identifying the best practices, challenges, and opportunities for improving the quantitative methods for accident risk level analysis in construction.
• Offering practical recommendations and implications for construction safety practitioners and policymakers based on the results of the systematic review.

2. Methodology

This section describes the research design and process of the systematic review, following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [14].

2.1. Data Sources and Search Strategy

The data sources for the systematic review are the following electronic databases: Scopus, Web of Science, and ProQuest, which cover the fields of construction, engineering, and safety. The search strategy for the systematic review is based on the following steps: (1) Identify the key terms and synonyms related to construction accidents, risk analysis, and quantitative methods. (2) Combine the key terms and synonyms using Boolean operators (AND, OR, and NOT) and truncation to create a comprehensive and sensitive search string. The detailed search strings for each database, along with their respective Boolean combinations, are presented in

Table 1. Search strategy for the systematic review.

Database	Search String
Scopus	TITLE-ABS-KEY (“construction*” OR “building*” OR “project*”) AND TITLE-ABS-KEY (“accident*” OR “incident*” OR “injury*” OR “fatality*”) AND TITLE-ABS-KEY (“risk*” OR “hazard*” OR “danger*”) AND TITLE-ABS-KEY (“analysis*” OR “assessment*” OR “evaluation*” OR “measurement*” OR “prediction*” OR “optimization*”) AND TITLE-ABS-KEY (“quantitative*” OR “statistical*” OR “mathematical*” OR “simulation*” OR “artificial intelligence*”) AND TITLE-ABS-KEY (“method*” OR “model*” OR “technique*” OR “tool*”)
Web of Science	TS=(“construction*” OR “building*” OR “project*”) AND TS=(“accident*” OR “incident*” OR “injury*” OR “fatality*”) AND TS=(“risk*” OR “hazard*” OR “danger*”) AND TS=(“analysis*” OR “assessment*” OR “evaluation*” OR “measurement*” OR “prediction*” OR “optimization*”) AND TS=(“quantitative*” OR “statistical*” OR “mathematical*” OR “simulation*” OR “artificial intelligence*”) AND TS=(“method*” OR “model*” OR “technique*” OR “tool*”)
ProQuest	AB((“construction*” OR “building*” OR “project*”) AND (“accident*” OR “incident*” OR “injury*” OR “fatality*”) AND (“risk*” OR “hazard*” OR “danger*”) AND (“analysis*” OR “assessment*” OR “evaluation*” OR “measurement*” OR “prediction*” OR “optimization*”) AND (“quantitative*” OR “statistical*” OR “mathematical*” OR “simulation*” OR “artificial intelligence*”) AND (“method*” OR “model*” OR “technique*” OR “tool*”))

Note: The search strings are tailored to the syntax of each database. Boolean operators (AND, OR, and NOT) are used to combine key terms, while truncation (*) allows for the inclusion of various word forms.

.

2.2. Inclusion and Exclusion Criteria

The inclusion and exclusion criteria for selecting relevant studies for the systematic review are based on the PICOS framework, which stands for population, intervention, comparator, outcome, and study design [15]. The criteria are as follows:

• Population: The population of interest is construction workers or construction projects in any country or region.
• Intervention: The intervention of interest is any quantitative method for accident risk level analysis in construction, such as statistical analysis, mathematical modeling, simulation, and artificial intelligence.
• Comparator: The comparator of interest is any other quantitative method for accident risk level analysis in construction, or no intervention (baseline or control).
• Outcome: The outcome of interest is the accident risk level or any related indicators, such as accident frequency, severity, probability, impact, etc.
• Study design: The study design of interest is any empirical study that applies a quantitative method for accident risk level analysis in construction, such as observational, experimental, quasi-experimental, or mixed-method studies.

The exclusion criteria are studies not published in English, not peer-reviewed, not accessible in full text, reviews, conference papers, editorials, opinions, commentaries, studies not published between 2013 and 2023, studies that do not report a quantitative method for accident risk level analysis in construction, and studies that do not report the accident level or any related indicators as an outcome.

2.3. Study Selection

The study selection for the systematic review is presented in Figure 1. The study selection for the systematic review is based on the following steps:

• Search results from each database are imported into EndNote (version 21), a reference management software chosen for its robust capabilities in organizing and managing large volumes of references, as well as its efficient duplicate removal features [16].
• Titles and abstracts of the search results using the inclusion and exclusion criteria are screened, and irrelevant studies are excluded.
• The full texts of the remaining studies are retrieved and screened using the inclusion and exclusion criteria, and ineligible studies are excluded.

The initial search yielded a total of 7810 articles from the databases. The distribution of articles from each database was as follows: Scopus: 5596 articles, Web of Science: 1553 articles, and ProQuest: 661 articles. After removing duplicates, 3804 articles remained. These articles were screened based on their titles and abstracts, resulting in 320 articles being selected for further review. The full texts of the 320 articles were assessed against the inclusion and exclusion criteria. This step resulted in 81 articles being deemed relevant for the systematic review.

2.4. Review Process Limitations

First, despite the execution of extensive literature searches, it is possible that selection bias may have influenced the inclusion of studies. Some studies may have unintentionally been excluded due to discrepancies in search terms or limitations in database coverage, which could potentially affect the generalizability of the findings. Second, the review may be subject to publication bias, as it only includes studies published in peer-reviewed journals. Relevant studies published in other formats, such as technical reports or gray literature, may have been excluded, leading to potential bias in the findings.

3. Overview of the Literature

3.1. Results of Year of Publication

The distribution of articles on quantitative accident risk level analysis in the construction industry by year of publication is displayed in Figure 2. It indicates the number of papers that were published in each year from 2013 to 2023. The number of papers rose from 2 in 2013 to 19 in 2023, indicating that the research topic has garnered increasing interest and popularity in recent years. The peak year of publication was 2023, with 19 papers, followed by 2022, with 15 papers. This growth is attributed to several factors, including technological advancements, which have enabled more effective risk analysis and safety measures through the rise of data analytics and machine learning [17]. Additionally, new safety regulations have prompted more rigorous research to ensure compliance with evolving standards [17]. There has also been heightened safety awareness, driven by increased scrutiny from the public and industry stakeholders, which has led to more research on risk management [18]. Furthermore, the industry faces emerging challenges related to sustainability and workforce diversity, creating new opportunities for research [18]. However, the decline in publications in 2020 and 2021 is largely due to the COVID-19 pandemic, which disrupted research and data collection efforts as many researchers shifted their focus to pandemic-related topics [19]. Overall, the upward trend in publications indicates a dynamic response to evolving challenges and a commitment to improving safety in construction.

3.2. Results of Publications by Country/Region

Figure 3 shows the distribution of the papers on quantitative accident risk level analysis in construction by country or region. It indicates the number of papers that were published or conducted in each country or region from 2013 to 2023. This review focused on the research location, as opposed to other studies that examined the publications according to the countries of the authors or institutions [20,21]. The United States, China, and South Korea were the most active and productive countries or regions in this research topic, accounting for 60.49% of the total papers. Also, the research topic was of global interest and relevance, as it involved papers from different continents and regions, such as Europe, Asia, Africa, and Oceania.

3.3. Results of Publication by Journal

Table 2. Overview and distribution of journal by publications.

Journal Title	Impact Factor	Journal Citation Ranking (%)/Quantile Rankings	Number of Papers
J Const Eng Manag	4.1	17.13/Q1	22
Saf Sci	4.7	19.81/Q1	21
Autom Constr	9.6	0.55/Q1	11
Buildings	3.1	28.18/Q2	6
J Saf Res	3.9	3.04/Q1	4
J Manag Eng	5.3	10.5/Q1	3
J Build Eng	6.7	4.97/Q1	2
J Civ Eng Manag	4.3	14.36/Q1	2
KSCE J Civ Eng	1.9	50.28/Q3	2
Reliab Eng Sys Saf	9.4	3.77/Q1	2
Risk Anal	3.0	13.43/Q1	2
IEEE T Autom Sci Eng	5.9	16.67/Q1	1
J Comput Civ Eng	4.7	11.60/Q1	1
Sci Rep	3.8	18.66/Q1	1
Struct Saf	5.7	8.84/Q1	1

and Figure 4 show the distribution of the papers on quantitative accident risk level analysis in construction by journal title. It indicates the number of papers that were published in each journal from 2013 to 2023. As shown in Table 2, the research papers are published across 15 different journals, with the majority of these journals ranking at the Q1 level. This high ranking reflects the rigorous peer-review processes and the credibility of the journals, thereby guaranteeing the overall quality and reliability of the papers which reviewed. The inclusion of predominantly Q1 journals further validates the significance and impact of the research within the academic community.

The analysis reveals that the most popular and influential journals for this research topic were the Journal of Construction Engineering and Management and Safety Science, with 22 and 21 papers, respectively. Figure 4 also shows that the research topic is relevant and interdisciplinary, as it involves papers from different fields and disciplines, such as engineering, management, automation, reliability, and risk. The research topic was aligned with the scope and aims of the journals, as well as the interests and preferences of the authors and readers.

3.4. Results of the Terms Analysis

The term analysis using VOSviewer (version 1.6.20) was performed to identify and visualize the main topics and themes of the literature on the methodologies employed for quantitative accident risk level analysis in construction [22]. The terms or keywords that appeared in the titles and abstracts of the 81 selected papers were extracted and imported into VOSviewer. A co-occurrence network of the terms was created and explored using VOSviewer [22]. The network visualization of the terms, shown in Figure 5, displays a graphical representation of the relationships among the terms based on their co-occurrences in the selected literature. In this visualization, nodes represent individual terms, with their size correlating to the frequency of occurrence in the papers. Larger nodes indicate terms that are mentioned more frequently, suggesting a higher level of interest or relevance within the literature. Additionally, the colors of the nodes indicate clusters of related terms, with each color representing a specific thematic group. Links between nodes illustrate the co-occurrence relationships among terms, with thicker lines denoting stronger relationships. A thick line suggests that the connected terms often appear together in the same titles or abstracts, thereby highlighting their interrelatedness. The network visualization of the terms reveals the following findings: “fall” occurred 14 times with a relevance of 3.82, “fatal accident” occurred 10 times with 2.92 relevance, and “construction safety” occurred 13 times with 2.15 relevance. These findings underscore the focus of the literature on specific risks and safety measures within the construction industry.

4. Results

4.1. Methodological Approaches

The systematic review examined a broad spectrum of methodological approaches applied in quantitative accident risk level analysis within the construction sector. These approaches were thoroughly assessed and classified into distinct categories, including statistical analysis, mathematical modeling, simulation, and artificial intelligence, as shown in

Table 3. Categorization of methodological approaches.

Category	Method	Number of Papers	References
Statistical Analysis	Correlation analysis Clustering method Regression analysis Frequency analysis	34	[9,10,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54]
Mathematical Modeling	Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) model Fuzzy set theory Analytical hierarchy process (AHP) Fault tree analysis Structural equation modeling	24	[23,37,44,49,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73]
Simulation	Monte Carlo simulation Discrete event simulation (DES)	16	[10,33,74,75,76,77,78,79,80,81,82,83,84,85,86,87]
Artificial Intelligence	Machine learning algorithms (Random Forest, adaptive boosting (AdaBoost), XGBoost, etc.) Optimization algorithms (Genetic algorithm, Particle Swarm Optimization, etc.) Deep learning (Computer vision and natural language)	23	[23,24,32,38,59,66,71,76,78,88,89,90,91,92,93,94,95,96,97,98,99,100,101]

.

Various studies incorporated statistical analysis techniques, such as correlation analysis, Mann–Whitney U Test, ANOVA, and regression modeling. Notably, regression analysis was further subdivided based on its application context. Ordinary least squares regression and spatial regression modeling were classified under statistical analysis due to their frequent use for hypothesis testing and linear relationship modeling. However, these methods can fall short when addressing non-linear relationships and interactions among variables, potentially oversimplifying the complex dynamics of construction accidents. In contrast, polynomial regression was categorized under mathematical modeling, reflecting its use in constructing predictive models that capture non-linear patterns in risk factor interactions. These methods were utilized to investigate relationships between variables, evaluate differences between groups, and model the connection between risk factors and accident occurrence. While polynomial regression can offer a better fit for non-linear data, it also risks overfitting, particularly in smaller datasets. Additionally, techniques, such as trimmed mean and variance, including the Winsorized variance in Welch, Yuen, and bootstrapping methods, were applied to address outliers and ensure the reliability of statistical conclusions. Yet, the choice of these techniques can influence the interpretation of results, and the appropriateness of their application often hinges on the underlying data distribution.

Mathematical modeling techniques were widely employed in several studies, covering methods such as factor analysis, principal components analysis, and polynomial regression. These approaches aimed to quantify risk factors and their interactions, often utilizing scale normalization techniques, and establishing assessment functions based on the likelihood and severity of accidents. While these models can provide valuable insights, their reliance on the assumptions of linearity and normality can limit their applicability in real-world scenarios where data may not conform to these assumptions.

Simulation techniques were prevalent in the reviewed literature, with researchers employing methods such as the discrete event simulation (DES), Monte Carlo simulation, and two-dimensional cellular automation (CA) model to simulate complex construction environments, analyze the propagation of risk factors, and assess the effectiveness of safety interventions. Furthermore, DES models facilitated the analysis of incident frequencies, durations, and severities, while the Monte Carlo simulation provided probabilistic estimates of accident risk levels under various scenarios. While these methods allow for a comprehensive exploration of potential accident scenarios, they also require significant data input and computational resources, which can be a barrier for practical implementation in the field. Furthermore, the assumptions made in simulation models can lead to results that may not accurately reflect real-world conditions.

A wide range of artificial intelligence techniques was also utilized in the reviewed studies, including machine learning algorithms, such as Random Forest, Support Vector Machine (SVM), Artificial Neural Network (ANN), and XGBoost for predictive modeling, risk classification, and feature selection. These methods leveraged large datasets comprising accident records, safety inspections, and environmental conditions. While machine learning techniques have shown great promise in identifying patterns and making predictions, they can be sensitive to the quality of the input data, leading to biased outcomes if the training data are not representative of the broader context. Additionally, techniques, such as genetic algorithms and Bayesian networks, were employed to optimize risk assessment models, identify influential risk factors, and enhance decision making in construction safety management.

The review identified a plethora of methodological approaches employed in quantitative accident risk level analysis within the construction industry. From conventional statistical methods to advanced machine learning algorithms and pioneering modeling approaches, researchers have utilized a variety of techniques to improve their comprehension of construction safety trends. Each approach offered unique insights into the complex dynamics of construction safety, showcasing the diverse strategies adopted by researchers to assess and mitigate accident risks effectively. However, a critical assessment of these methodologies highlights the need for ongoing refinement and adaptation to ensure their relevance and applicability in real-world construction safety management.

4.2. Data Sources

Table 4. Results of data sources.

Data Source	Number of Papers
Data from national institutions	34
Data from industrial companies	15
Experimental data/simulation data/data from sensors	15
Focus group interview/questionnaire survey	17
Total	81

presents the data sources employed in the quantitative assessment of accident risk levels within the construction industry. A prominent source of information in the reviewed literature was national-level datasets, with 34 papers utilizing data sourced from governmental agencies, regulatory bodies, and national statistical databases. These datasets provided comprehensive insights into accident rates, trends, and patterns across the construction sector, facilitating the macro-level analysis of accident risk levels. Fifteen papers drew upon data directly obtained from construction companies, including accident records, safety reports, and internal databases. These studies often focused on specific construction projects or organizations, allowing for a detailed analysis of site-specific risk factors, safety practices, and incident frequencies. A significant portion of the reviewed papers (15) relied on experimental data, simulation data, or data collected from sensors deployed in construction environments. These datasets facilitated the analysis of safety interventions, the simulation of accident scenarios, and the real-time monitoring of environmental conditions, contributing to a deeper understanding of accident risk dynamics. Seventeen papers utilized qualitative data obtained through focus group interviews, questionnaires, or surveys conducted with construction workers, safety professionals, and other stakeholders. These studies explored subjective perceptions of risk, safety culture, and behavioral factors influencing accident occurrences, complementing quantitative analysis with qualitative insights.

4.3. Comparative Analysis

The comparative analysis conducted in this systematic review aimed to elucidate the differences and similarities among the various methodological approaches employed in quantitative accident risk level analysis within the construction industry. The review encompassed a comprehensive examination of the effectiveness, applicability, and limitations of each approach, shedding light on their respective contributions to advancing construction safety management.

4.3.1. Statistical Analysis vs. Mathematical Modeling

Statistical analysis techniques, including correlation analysis and ANOVA, were frequently utilized to explore relationships between risk factors and accident occurrence. These methods offered valuable insights into the statistical significance of variables and the identification of potential predictors of accidents. Conversely, mathematical modeling techniques, such as factor analysis and polynomial regression, provided a deeper understanding of the underlying mechanisms driving accident risk. By quantifying risk factors and their interactions, mathematical models facilitated the development of predictive models and risk assessment frameworks, enhancing the accuracy and precision of risk analysis in construction.

4.3.2. Simulation vs. Artificial Intelligence

Simulation-based approaches, such as discrete event simulation and Monte Carlo simulation, enabled researchers to simulate complex construction environments and analyze the propagation of risk factors over time. These methods offered a dynamic perspective on accident risk, allowing for the assessment of risk dynamics and the evaluation of the effectiveness of safety interventions. In contrast, artificial intelligence techniques, including machine learning algorithms like Random Forest and Support Vector Machine, focused on predictive modeling and risk classification. By leveraging large datasets and advanced algorithms, artificial intelligence methods provided robust predictive capabilities, identifying patterns and trends in accident occurrence and facilitating proactive risk management strategies.

4.3.3. Statistical Analysis vs. Simulation

While statistical analysis techniques were effective in exploring relationships between variables and identifying statistical associations, they were often limited in their ability to capture the dynamic nature of construction environments. Simulation-based approaches, on the other hand, provided a more holistic view of accident risk by simulating complex interactions among risk factors and evaluating their impact on accident occurrence. By incorporating temporal and spatial dynamics, simulation methods offered insights into the evolution of risk over time and space, enabling stakeholders to proactively identify and mitigate potential hazards.

4.3.4. Artificial Intelligence vs. Mathematical Modeling

Artificial intelligence techniques, characterized by their data-driven approach and predictive capabilities, complemented mathematical modeling methods by offering advanced predictive modeling capabilities. While mathematical models provided insights into the underlying mechanisms driving accident risk, artificial intelligence algorithms leveraged large datasets to identify complex patterns and trends in accident occurrence. By combining the strengths of both approaches, researchers were able to develop robust predictive models that accounted for both deterministic and stochastic factors influencing accident risk in construction.

4.4. Results of Risk Level Analysis

The systematic review yielded a comprehensive set of results pertaining to the analysis of accident risk levels within the construction industry. The findings, derived from diverse methodological approaches, data sources, and analytical techniques, provided valuable insights into the prevalence, predictors, and implications of accident risks in construction projects.

4.4.1. Prevalence of Accident Risks

Across the reviewed studies, there was a consistent recognition of the substantial prevalence of accident risks within the construction industry. Statistical analyses, including frequency rate calculations and correlation analyses, consistently highlighted the elevated incidence of accidents and injuries compared to other sectors [25,34,39,43,50,87]. The fatality rate, calculated as the number of fatalities per 100,000 full-time equivalent workers [9] or number of fatalities per 10,000 full-time equivalent workers [10], underscored the severity of accidents in construction settings. Furthermore, simulation-based approaches, such as Monte Carlo simulation and discrete event simulation models, enabled the estimation of accident frequencies and severities, emphasizing the dynamic nature of risk within construction environments [10,74,81,86]. According to Jeong et al.’s (2022) results, the highest number of fatal incidents occurred during reinforced concrete work. The highest rate of fatalities per 10,000 workers was seen in scaffolding and demolition work, at 28.90%. Scaffolding and demolition works were the most dangerous work, with an average normalized ranking (ANR) of 2.51. The ANR classified 27 construction works into five groups based on the risk ratings, which ranged from 1 to 5 [10]. Results from Gernard (2022) indicated that, in the years 2020–2029, workers in the solar PV installation industry are expected to have a mean of 16.6 nonfatal recordable safety incidents [86]. Project cost overruns and safety risk incidents are strongly correlated, according to Alkaissy et al.’s (2022b) findings. Consequently, when it comes to project cost management, safety concerns in the resource pool should be addressed. Additionally, the results highlighted the important relationship between safety risk incidents and various age groups in the workforce. It also came to light that the middle-aged group’s (i.e., 40–49-year olds) compensation claims had a major influence on project cost overruns [87].

4.4.2. Predictors of Accident Risks

Several risk factors were identified as significant predictors of accident risks in construction projects. Machine learning algorithms, including Random Forest, Support Vector Machine (SVM), and Artificial Neural Network (ANN), facilitated the identification and classification of influential risk factors based on large-scale datasets [24,88,89,90,91,94,97,98,99,100,101]. Factors such as working days lost, type of injury, worker-related information (age, daily wage, and experience), worker behaviors, environmental conditions, project complexity, and organizational culture emerged as key determinants of accident risks. Moreover, mathematical modeling techniques, such as factor analysis and principal components analysis, enabled the quantification and prioritization of risk factors, elucidating their respective contributions to overall risk levels [58,67]. Findings from the post-accident disability status prediction model developed by Koc et al. (2021) showed that XGBoost could identify 67.74% of all workers with permanent disabilities, yielding an accuracy of 0.8292 [88]. Results from Koc et al.’s (2023) accident outcome prediction model had an accuracy of 0.6675 and were able to detect 792 fatal incidents correctly. Also, findings from the game theory-based Shapley additive explanations (SHAP) analysis showed that company (such as the number of past accidents and workers in the company) and worker-related (such as age, daily wage, experience, shift, and past accidents of the workers) attributes were the most influential factors in accident outcome (fatal or nonfatal) [89]. According to a study by Choi et al. (2020), the Random Forest method has a 91.98% success rate in identifying which workers will be at risk of loss of life. The study’s Random Forest analysis reveals that month (season) and employment size are the most significant factors, followed by age, weekday, and service length, based on mean decrease Gini values to estimate the likelihood of a fatal accident [94].

4.4.3. Implications of Accident Risks

The analysis of accident risk levels unveiled significant implications for construction safety management and policy development. Simulation-based approaches, such as fault tree analysis and multivariate joint distribution models, elucidated the cascading effects and interdependencies of risk factors, highlighting critical control points and vulnerabilities within construction processes [44,57]. Additionally, machine learning algorithms, coupled with risk quantification methods, such as the Likelihood, Consequences, and Exposure (LCE) model, provided actionable insights into prioritizing safety interventions and allocating resources effectively [26]. Furthermore, Bayesian network models and decision matrix risk assessment techniques offered probabilistic estimates of accident probabilities and severities, enabling informed decision making under uncertainty [38,71,78,92,93].

4.4.4. Overall Assessment and Synthesis

The synthesis of results underscored the multifaceted nature of accident risk analysis in construction, encompassing a wide spectrum of methodologies, data sources, and analytical approaches. While statistical analyses provided valuable descriptive insights into accident prevalence and severity, simulation and modeling techniques offered predictive capabilities and scenario-based risk assessments. Machine learning algorithms, in particular, emerged as powerful tools for identifying complex patterns and interactions among risk factors, facilitating proactive risk management strategies. However, the review also revealed methodological challenges, such as data quality issues and model validation concerns, underscoring the need for further research and refinement of analytical frameworks.

4.5. Strengths and Limitations

The systematic review unveiled several strengths and limitations inherent in the methodologies utilized for quantitative accident risk level analysis within the construction industry. The strengths include the following:

• Objective and Reliable Results: One of the primary strengths observed across the reviewed studies was the ability to provide objective and quantitative assessments of accident risk levels. The utilization of quantitative methodologies, such as statistical analysis, mathematical modeling, simulation, and artificial intelligence, facilitated the generation of objective and reliable results. These methods enabled researchers to quantify risk levels, predict accident probabilities, and identify influential risk factors based on empirical data without subjective bias or influence, using numerical or measurable values and criteria [78,93].
• Predictive Capabilities: Many of the reviewed studies demonstrated strong predictive capabilities, particularly those utilizing simulation models and machine learning algorithms, such as Random Forest, Support Vector Machine (SVM), and Artificial Neural Network (ANN). These techniques effectively forecasted accident occurrences, contributing to proactive risk management and prevention strategies [23,98,100].
• Comprehensive Risk Assessment: The incorporation of diverse data sources, including accident records, safety inspections, and environmental conditions, enabled a comprehensive assessment of accident risk levels. This multidimensional approach provided a holistic understanding of the factors influencing construction safety, facilitating targeted interventions [51,53,73].
• Optimization of Risk Mitigation Strategies: Several methodologies, such as genetic algorithms and game theory-based Shapley Additive Explanations (SHAP), focused on optimizing risk mitigation strategies [24,77,88,89]. By identifying critical risk factors and their interactions, these approaches informed the development of tailored interventions to minimize accident risks effectively.
• Integration of Advanced Technologies: This review highlighted the integration of advanced technologies, such as artificial intelligence and simulation, into accident risk level analysis. These technologies enhanced the accuracy and efficiency of risk assessment models, enabling real-time monitoring, scenario analysis, and decision support in construction safety management [23,74,98,100].
• Integration of Expert Knowledge: Many studies integrated expert knowledge and domain expertise into quantitative risk assessment frameworks, enhancing the accuracy and reliability of the results. Techniques such as the analytic hierarchy process (AHP) and focus group discussions facilitated the elicitation and incorporation of stakeholders’ perspectives, ensuring comprehensive risk evaluations that align with industry practices and regulatory requirements [34,59,60,63,65].

The limitations include the following:

• Data Availability and Quality: A notable limitation observed across several studies was the challenge associated with data availability and quality. The reliability of quantitative risk assessments heavily relied on the availability of accurate and comprehensive data sources, including accident records, safety inspections, and environmental data. Inadequate or incomplete datasets often hindered the effectiveness of risk analysis models and led to potential biases in the results [55,91]. Especially in developing countries or regions, it is often difficult or costly to obtain large amount of data to perform accident risk level analysis [29].
• Complexity of Construction Environments: The reviewed methodologies often struggled to capture the complex and dynamic nature of construction environments. The challenges posed by factors such as evolving site conditions, human behavior, and organizational factors complicated the assessment of accident risk levels and the implementation of risk mitigation strategies [58,78,89].
• Methodological Assumptions: Many methodologies relied on certain assumptions and simplifications, which may not always reflect the complex realities of construction environments. Assumptions related to the independence of variables, linear relationships, and stationary conditions could impact the validity and generalizability of results.

5. Discussion

5.1. Research Gaps and Opportunities

The systematic review revealed considerable gaps in the current literature on quantitative accident risk level analysis in construction, offering promising opportunities for future research and methodological advancements.

Gaps in the current literature includes:

(1)

Limited integration of emerging technologies: Although some studies incorporated emerging technologies, like artificial intelligence and machine learning, there is a notable gap in the comprehensive integration of these technologies into quantitative risk analysis frameworks. Future research could explore innovative applications of technologies, like drones, sensors, digital twins, and virtual reality, for real-time risk assessment and proactive hazard identification.

(2)

Limited focus on human factors: While some methodologies considered workers’ behaviors and cognitive factors, there is a gap in the comprehensive integration of human factors into quantitative risk analysis frameworks. Future research could explore interdisciplinary approaches that incorporate insights from psychology, sociology, and human factors engineering to better understand and mitigate human-related risks in construction. Additionally, these human factors could be integrated into existing quantitative models to provide a more comprehensive risk assessment.

(3)

Need for standardized risk assessment frameworks: The review identified a lack of standardized frameworks for quantitative accident risk level analysis in construction, leading to inconsistencies in methodologies and metrics across studies. Future research could focus on developing standardized risk assessment frameworks that encompass a broad range of risk factors, ensure comparability between studies, and facilitate evidence-based decision making in construction safety management.

5.2. Recommendations for Future Research

Based on the outcomes of the systematic review, several recommendations for future research in the field of quantitative accident risk level analysis in construction emerge, aimed at addressing existing gaps and advancing the state of the art in construction safety management.

• Firstly, future research should explore the integration of advanced technologies, such as artificial intelligence, machine learning, and real-time sensing systems, to enhance the accuracy and efficiency of quantitative risk assessment models. Investigating the feasibility and effectiveness of incorporating wearable devices, Internet of Things (IoT) sensors, and unmanned aerial vehicles (UAVs) into risk analysis frameworks could significantly improve the real-time monitoring and prediction of accident risks on construction sites.
• Secondly, there is a need for the development of comprehensive risk assessment frameworks that consider multiple dimensions of risk, including physical hazards, human factors, and organizational factors. Future research should focus on integrating these diverse risk factors into unified risk assessment models, utilizing advanced analytical techniques such as fuzzy logic, Bayesian networks, and multi-criteria decision-making methods to capture the complexity of construction safety dynamics.
• Thirdly, the validation and benchmarking of risk assessment models against real-world accident data are essential for assessing their reliability, accuracy, and generalizability. Future research should prioritize validation studies using longitudinal accident datasets from construction projects of varying scales and contexts. Comparative studies evaluating the performance of different risk assessment models under diverse conditions can provide valuable insights into their strengths, limitations, and applicability in practice.
• Fourthly, enhancing the quality, accessibility, and interoperability of construction safety data is critical for advancing quantitative risk analysis in the field. Future research should concentrate on standardizing data collection protocols, creating interoperable data repositories, and leveraging emerging technologies, such as blockchain and distributed ledger technology (DLT), to enhance data integrity and transparency. Additionally, efforts to enhance data sharing and collaboration among stakeholders can facilitate the creation of comprehensive accident databases for research and practice.
• Fifthly, the empirical evaluation of safety interventions is essential for assessing their effectiveness in mitigating accident risks and improving construction safety outcomes. Future research should prioritize rigorous evaluation studies that measure the impact of safety interventions on accident rates, near-miss incidents, and safety culture indicators. Employing experimental designs, randomized controlled trials, and quasi-experimental methods can provide robust evidence of intervention effectiveness and inform evidence-based safety practices in the construction industry.
• Finally, investing in education and training initiatives is crucial for building a skilled workforce equipped with the knowledge and competencies to implement quantitative risk analysis methodologies effectively. Future research should focus on developing innovative educational programs, training modules, and certification courses that integrate quantitative risk analysis principles into construction management curricula. Promoting the awareness and adoption of advanced risk assessment tools and techniques among construction professionals can foster a culture of safety excellence and continuous improvement in the industry.

Overall, these recommendations provide a roadmap for future research endeavors aimed at advancing the field of quantitative accident risk level analysis in construction. By addressing these research priorities, researchers, practitioners, and policymakers can collectively contribute to enhancing construction safety management practices and ultimately reducing the incidence of accidents and injuries in the construction industry.

While this systematic review does not present a single, groundbreaking solution, it offers a critical analysis of the current landscape of quantitative risk quantification methods in construction safety management. By identifying significant gaps in areas like technology integration, human factor consideration, and standardization, this research lays the groundwork for future advancements in this field. Addressing these gaps presents exciting opportunities to develop more robust, data-driven, and human-centered risk quantification frameworks. The potential applications of these advancements are far-reaching, with the potential to improve accident risk prediction accuracy, enable proactive hazard identification, and empower construction professionals with better decision-making tools. Ultimately, this research paves the way for future endeavors that can significantly contribute to a safer construction industry with fewer accidents and injuries.

6. Conclusions

• We conducted a systematic review of methodologies for quantitative accident risk level analysis in the construction industry, following the PRISMA guidelines.
• We identified a diverse range of approaches:

  ○

  Statistical Analysis: Explored relationships between risk factors and modeled likelihood and severity of accidents.

  ○

  Mathematical Modeling: Quantified risk factors and their interactions, providing insights into accident dynamics.

  ○

  Simulation-Based Methods: Simulated complex environments to analyze risk propagation and evaluate safety interventions.

  ○

  Artificial intelligence: Enhanced predictive modeling, feature selection, and decision support through machine learning and optimization techniques.

• We highlighted the importance of a multifaceted approach to effectively assess and mitigate accident risks in construction projects.
• We emphasized challenges, such as data quality, methodological constraints, and the need for collaboration among researchers, practitioners, and decision makers, to improve safety management practices.
• We offered guidance for enhancing risk assessment processes and improving safety outcomes in the construction industry.
• We suggested future research should refine methodologies and leverage emerging technologies to advance construction safety management.