A Quantitative Analysis of Decision-Making Risk Factors for Mega Infrastructure Projects in China

Abstract

The “trillion-dollar era” of megaprojects has increased the demand for the scope of mega infrastructure. To address the requirement for high-quality “investment, construction, and operation” integration, the EPC and PPP models must be combined. The complexity of megaprojects has resulted in the complexity of project risk variables under the new model. However, few existing studies have undertaken in-depth studies on the risk of EPC + PPP megaprojects. The interplay and dynamic evolution of risk factors, in particular, have not been taken into account. This research intends to fill this gap by systematically identifying and modeling the risk variables associated with the adoption of the EPC + PPP model for mega infrastructure projects. In this study, the Bayesian network is used to detect decision-making risk variables for large infrastructure projects in China. The findings indicate that (i) 22 influencing factors of megaproject decision making are identified, including organizational decision making, PPP investment and financing, EPC construction, operations management, and policy and law. (ii) Considering the real project decision-making process, a model based on a dynamic Bayesian network is built, and associated governance techniques and early warning protection mechanisms are designed for the decision-making process. (iii) Using the Yiwu Mall Avenue project as an example, the Bayesian simulation model of decision-making risks is applied to a typical case to validate its feasibility and correctness. These findings have significant theoretical and practical implications for research on the identification and governance of decision-making risks in megaprojects using the EPC + PPP model in China.

1. Introduction

Mega infrastructure projects are a crucial component of the support for social and economic growth. To maintain the necessary global economic growth, infrastructure costs will exceed USD 94 trillion in 2040, according to the report [1] released by the Global Infrastructure Hub. An additional USD 3.5 trillion will be required to achieve the Sustainable Development Goals (SDGs) aim of reducing global household water and electricity use by 2030. Hence, the cumulative global infrastructure requirements are projected to amount to the staggering sum of USD 97 trillion. Infrastructure projects have the duty and obligation to generate wider social value beyond the primary advantages derived from constructing assets [2]. Consequently, mega infrastructure projects will be designed with more emphasis on their convergence patterns [3]. Furthermore, scholars express apprehension regarding the sustainability of mega infrastructure projects, owing to their extensive scope, substantial financial commitment, intricate technical nature, and profound societal implications [4].

Currently, there is a pressing need for mega infrastructure projects to thoroughly examine the path towards value addition and to reconstruct the value associated with public goods. The Public–Private Partnership (PPP) model and the Engineering Procurement Construction (EPC) model individually fail to adequately address the requirements of achieving high-quality integration goals in “investment–build–operation” projects. Consequently, a pragmatic solution to this predicament lies in the integration of the EPC and PPP models. Their comprehensive incorporation in mega infrastructure endeavors aligns with the construction contracting market and has the potential to foster innovation in the management practices of the construction sector. The subject matter possesses prospective scholarly significance and practical applicability [5]. Nevertheless, the use of the EPC + PPP model has certain inherent risks regarding policy, contract, construction, and operation. The existing literature and empirical evidence indicate that megaprojects are characterized by their inherent complexity, making it challenging to establish clear decision-making objectives. Furthermore, these projects operate within dynamic and evolving implementation environments [6], which further complicates the decision-making process. Additionally, decisionmakers often possess limited knowledge and understanding of the project, further exacerbating the complexity of the decision-making process [7]. The numerous sources of risk variables and their potential for irreversible incidents, as exemplified by the Rakum Grand Canal project in the former Soviet Union and the overturning of the water stop during the installation of the Busan Geoje Immersed Tube Tunnel, contribute to the complexity of decision-making processes [8,9]. The identification of risk factors associated with decision making and the thorough examination of their underlying mechanisms are crucial requirements for enhancing the effectiveness of risk management and promoting evidence-based decision making in large-scale initiatives [10].

To mitigate the aforementioned constraints, this research endeavor serves as an exploratory investigation into the realm of risk associated with decision making in megaprojects. This study centers on the theory and application of risk management in the context of decision making for megaprojects. The primary aim of this study is to address the following research questions: (i) How can quantitative methods be used to identify risk factors in mega infrastructure projects under the EPC + PPP model? (ii) How can a five-dimensional framework system of risk factors for megaprojects be constructed? (iii) How can the results of this study be translated into actionable insights to advise decisionmakers on project management strategies? Hence, this study employs Bayesian network theory to investigate the identification and management of risks in megaproject decision making inside the EPC + PPP model. Additionally, the simulation of the dynamic evolution of risk factors during the project construction process enhances the existing body of knowledge in project risk management theory.

2. Literature Review

2.1. Risk Studies on Government-Funded Projects

The foundation of risk management is in the identification, recognition, and categorization of risks. According to Furlong et al., it is posited that the dependability of national government funding will progressively diminish in the forthcoming years. Consequently, it is suggested that risk management procedures should be further enhanced to ensure the protection of project construction financing by expanding the customer base [11]. According to Gebre et al., the government is currently facing financial limitations in allocating funds towards road projects, leading to inadequate investment in road infrastructure. The facilitation of the execution of the PPP model for the project will be significantly influenced by this element [12]. Liu et al. analyzed the economic aspect of the primary engineering risks associated with China’s outward investment. They highlighted that the limitations and diverse measures imposed by developed nations on China, such as taxation, industrial policies, and market regulations, can significantly impede the progress of construction projects [13]. Zayed et al. employed a hierarchical analysis approach to assess the risk and uncertainty associated with road developments in China. The risk elements associated with road projects were categorized into two main types: macrorisks and microrisks. To assess the current risk state of the project and aid the contractor in making prompt and suitable decisions, a risk index was computed [14].

2.2. Research on Risk Management Methods for Megaprojects

At this stage, there is limited research from domestic and foreign scholars on the risk management of megaprojects under the combined EPC + PPP model. The previous studies can serve as a reference for future research directions. Carbonara et al. identified key risk factors in infrastructure projects and produced a list of significant risks in two broad categories: demand/usage risks and cost overrun/financial closure risks [15]. Ozdoganm and Birgonu proposed six main types of risks that need to be considered during the planning phase of a PPP project. These include political, market, financial, legal and regulatory, construction, and operational risks. To assist in decision making, the authors have developed a framework for supporting PPP project decision-making processes [16]. Iyer and Sagheer identified 17 risks, such as legalistic risk, cost overrun risk, schedule overrun risk, etc., by analyzing road PPP projects in India [17].

The potential hazards associated with megaprojects implemented under the EPC and PPP model permeate the entirety of the project’s lifespan, and the elements contributing to these risks are intricate and subject to change. Hence, scholars employ Monte Carlo simulation [18], Fuzzy Comprehensive Evaluation [19], and Gray Correlation [20] methodologies to evaluate the identified risk variables, aiming to integrate qualitative and quantitative evaluations and provide a more intuitive depiction of the level of risk in the projects. Based on this premise, some scholars suggest approaches to ascertain shared risk through the integration of hybrid fuzzy techniques and Cybernetic Analysis Network Process (CANP) models [21]. Pythagorean Fuzzy Sets (PFSs) are employed to examine the allocation of risk between the government and investors [22]. Furthermore, Kardes et al. proposed a classification of project risks, distinguishing between internal and external risks [23]. Internal risks encompass factors such as organizational, financial, and personnel-related issues, while external risks pertain to political, subcontractor-related, and legal factors [24]. Cifrian et al. proposed a categorization of the PPP model into five distinct stages. It has been noted that infrastructure projects encompass two distinct categories of hazards: internal system risks and external environment risks [25].

Within the realm of risk assessment in engineering projects, researchers have extensively examined many approaches for evaluating risk. In their study, Hong et al. developed a comprehensive network diagram to analyze the various aspects that impact the advancement of home-building projects [26]. Miller et al. employed a hierarchical analysis approach to examine the attributes of the project. They established a set of criteria for evaluating the documents provided by the general contractor and sought the input of domain experts to assess and appraise these criteria [27]. Gordon et al. conducted a comprehensive analysis of the causes of operational risks leading to the failure of metro projects under the PPP model. They specifically examined a case study of a metro system that was entirely separate from its design, construction, and operations [28]. Ameyaw employed the Fuzzy Comprehensive Evaluation Method (FCM) as a means of assessing the hazards associated with PPP projects in underdeveloped nations [29].

2.3. Bayesian Modeling Research Applicable to Risk Assessment

The intricate nature and inherent ambiguity of megaprojects give rise to a multitude of constituent elements, each of which engenders potential dangers. The aforementioned aspects hold significant importance in the process of identifying risk interrelationships and transmission mechanisms, as well as critical risk variables and linkages. Nevertheless, the majority of the aforementioned approaches fail to consider the analysis of risk interdependencies and solely focus on evaluating the influence of individual risk variables on the project as a whole. Currently, there exists a limited number of scholarly investigations about the assessment of risk associated with decision making in megaprojects within the unique framework of EPC combined with PPP. The evaluation methodologies utilized in previous studies predominantly consist of explanatory structural models, social network analysis, and other similar approaches. While these methodologies examine the interconnectedness of risks, they do not possess the capability to thoroughly assess the qualitative and quantitative aspects of all risk factors. Hence, it is imperative to integrate qualitative and quantitative assessments to conduct risk identification in a more extensive and scientifically rigorous manner. Simultaneously, the utilization of the Bayesian network approach in simulation allows for a more intuitive and scientifically grounded identification of the pivotal aspects that influence decision-making risk. This method offers valuable insights and guidance for the development of risk management strategies.

The introduction of Bayesian network analysis can be attributed to Thomas Bayes during the 18th century. As the field of statistics progressed, Bayesian approaches gained recognition and were highly regarded [30]. Pearl (1986) introduced a graphical network that relies on probabilistic inference and provided a precise elucidation of Bayesian networks [31]. A Bayesian network is a graphical model that is commonly employed to conduct probabilistic inference [32]. The conventional Bayesian model is limited to representing a singular time slice structure, and when confronted with very intricate situations, it fails to account for the temporal impact on the entire process. A dynamic Bayesian network (DBN) is a type of Bayesian network that integrates temporal information to effectively describe time series data and capture the temporal dependencies among nodes [33]. The DBN is presently employed in a diverse range of applications, such as crisis warning [34], defect identification and route analysis [35], and data mining [36]. A DBN is a preferable choice when it comes to monitoring and predicting the values of random variables. Additionally, a DBN can effectively reflect the state of the system at any given point in time [37,38]. The predominant area of investigation within Bayesian research is the examination of logical reasoning in conjunction with empirical data. Conversely, there exists a dearth of scholarly inquiries that specifically address the use of Bayesian methods in practical engineering contexts [39,40].

In conclusion, the extant literature has examined the hazards associated with megaprojects; however, these studies have not adequately addressed the unique circumstances surrounding the integration of EPC with the PPP model. Furthermore, previous research has neglected to examine the interplay and evolving nature of risk factors. Hence, it is imperative to undertake comprehensive investigation, examination, and research regarding the hazards associated with megaprojects in China. Consequently, this study initially develops a risk framework for megaprojects operating under the EPC + PPP model by conducting a comprehensive review of the existing literature. Subsequently, the framework is assessed through the analysis of questionnaires and statistical methods. The utilization of a DBN is employed to assess the risks associated with megaprojects, considering both dynamic factors and correlated aspects. In conclusion, the integration of game theory and mathematical modeling is utilized to develop a risk response decision model that takes into account the varying objectives of decisionmakers. This model facilitates a comprehensive study of significant risks and aids in making informed response decisions.

3. Methodology and Data Presentation

3.1. Decision-Making Risk Factor Analysis

The initial phase of model creation involves performing a comprehensive examination of the risk factors regarding the decision-making risk model. This research undertakes a comprehensive evaluation of the pertinent literature to identify and synthesize the many risk variables. Recognizing the challenge of precisely quantifying risk factors in megaprojects, scholars have undertaken several studies to identify and classify such factors. Krane et al. conducted a study on seven megaprojects and identified two primary categories of factors influencing project risk management: strategic and operational risks [41]. Abbasi et al. proposed five risk indicators, namely simulated occurrence probability of a risk and simulated local and global influence of a risk, as well as total risk loss and total risk propagation loss of a project [42]. Cha developed project risk scoring (PRS), and these factors are classified into seven categories, including schedule, budget, quality, safety, environmental, contractual, and management issues [43]. Aladag et al. utilized fuzzy hierarchical analysis to prioritize risks and identified that risk factors with significant impact comprise improper partner selection, inadequate relationships with the employer, and incompetent contractor selection [44]. Siraj et al. identified that typical risks in construction projects include unpredictable inflationary changes, design failures, and changes in governmental policies and laws [45]. Badawy et al. grouped 43 risk factors into four categories of risk factors, including the contract type, implementation of risk management processes, the contract cost, and total project duration [46]. Abd Karim et al.’s survey of risk factors designed 25 common risk factors that were categorized into five categories, including shortage of material, late deliveries of material, insufficient technology, poor quality of workmanship, and cash flow difficulties [47]. Accordingly, this study develops a comprehensive framework that includes 22 factors to assess the risk framework system for mega infrastructure projects in China. This framework is presented in

Table 1. Risk systems for mega infrastructure decisions.

Risk Factors	Factor Description	References
Product/Service prices A01	Fluctuations in product/service prices have resulted in a decrease in the realized benefits of the EPC + PPP project compared with the initially anticipated benefits.	Bayat [48]; Wang [49];
Managerial operational level A02	The presence of several identities within social capital parties might result in inadequate managerial performance or an inability to conform to established government protocols and rules, thus impeding the efficiency of project operations.	Krane [41]; Zhang [50];
Market demand A03	This pertains to the potential for disparities between market predictions and actual demand as a result of shifts in market demand caused by reasons beyond the scope of uniqueness risk.	Sanboskani [51];
Integration of the bidding procedures of the two parties into one A04	The PPP + EPC model requires social capitalists to assume investment and construction responsibilities. This is necessary because social capitalists often lack the capacity to invest in or construct certain aspects of the project.	Ghasemi [52];
Financing costs A05	Fluctuations in market interest rates and inflationary pressures contribute to elevated price levels, subsequently resulting in increased financing expenses.	Taghizadeh-Hesary [53]; Jokar [54];
Bidding competition A06	This encompasses bidding methods that are deemed unjust, unfair, and lacking transparency, as well as the provision of inadequate or intentionally misleading information regarding the bidding project.	Kreye [55]; Qiao [56];
Contract negotiation A07	The two assessments and one case lack comprehensive argumentation, depth, and exhibit inaccuracies and confusing delineations.	Wang [57]; Branconi [58];
Completeness of contract documents A08	The presence of incomplete contracts in the domains of investment, construction, operation, and maintenance, coupled with ambiguous delineation of rights, responsibilities, and benefits, along with inequitable contractual provisions, have given rise to challenges pertaining to contractual documentation.	Cha [43]; Chen [59]; Shaikh [60]; Almarri [61];
Quality of design solution A09	In the lump-sum package model, the validation of the design program’s cost compression and adherence to requirements is insufficiently supported by social capital.	Wuni [62];
Access to finance A10	The inability to obtain financing within the designated timeframe, attributed to factors such as social capital or local debt, leads to the withdrawal of social capital and the subsequent stagnation of the project.	Steffen [63]; Sinha [64]; Chowdhry [65];
Operating cost A11	Government-mandated enhancements in service standards, escalations in operating expenses, and various market environment elements contribute to the occurrence of operating cost overruns and revenue decreases.	Welde [66]; Shi [67];
Adaptation of laws and regulations A12	The term primarily pertains to the potential risk arising from the implementation, dissemination, modification, and reinterpretation of laws and regulations, as well as the insufficiency of pertinent legal frameworks.	Riley [68]; Rossler [69]; Ozgur [70]; Berezin [71]; Alahmadi [72];
Governance of the relationship between the social capitalist and the government A13	Within the framework of the PPP and Entrepreneurial Public–Private Cooperation (PPP + EPC) model, the social capitalist assumes several roles, possesses many viewpoints, and harbors diverse interests during interactions with the government.	Zheng [73]; Demissew [74]; Feng [75];
Project adaptation A14	The potential impact on the successful implementation of EPC + PPP projects should be taken into account in light of potential alterations in key government decisionmakers or the formulation of novel policies or economic development strategies.	Nakamura [76]; Wang [77];
Selection of consortium members A15	The process of selecting consortium members has not undergone thorough and comprehensive evaluation, including factors such as company reputation, economic strength, and performance competence.	Siraj [45]; Aloysius [78]; Wang [79]; Aznar [80]
Multiple identities of social capitalists A16	The Social Capital Party’s management capacity, organizational and coordination abilities, and overall project management proficiency are not commensurate with the complex role of being an investor, general contractor, and operator.	Aladag [44]; Toschi [81]; Teller [82];
Decision-making interventions by governments A17	The government’s delays in initiating the project can be attributed to various factors, including procedural irregularities, bureaucratic tendencies, limited operational experience and capacity, inadequate preparation, and information asymmetry.	Müller [83]; Janssen [84]; Hasan [85]; Barrutia [86];
Control structure A18	The absence of trust in government implementing agencies has resulted in a biased approach towards the design of the cooperation mechanism. This has hindered the social capital party’s ability to effectively fulfill its role as the project’s legal entity.	Deep [87]; Arshad [88]; Mohammadi [89];
Stakeholder needs A19	The stakeholder approach to the PPP + EPC model of social capital investment and construction portfolio model lacks comprehensive understanding, resulting in resistance and a failure to achieve optimal alignment with the diverse demands of stakeholders.	Alaloul [90]; Rafeh [91];
Project schedule delay A20	Insufficient readiness in the preliminary phase of the PPP initiative resulted in a halt in building progress. Subsequently, the project encountered delays during the construction phase, leading to its inability to meet the predetermined deadline and contractual period.	Xie [92]; Ahmed [93];
Construction cost overrun A21	The occurrence of construction cost overruns can be attributed to various factors, including escalations in raw material prices, labor expenses, and uncertainties arising from the preliminary design papers.	Cha [43]; Plebankiewicz [94]; Adam [95]; Sovacool [96];
Engineering quality issues A22	Social capital has the dual role of investor and design–build contractor. This situation hinders the establishment of effective supervision and constraints on the contractor’s construction process, ultimately resulting in quality issues.	Abd Karim [47]; Ugural [97]; Ding [98];

.

3.2. Model Development

During the stage of model development, the determination of the number, type, and state of the nodes in the model is achieved through the processes of structural learning and parameter learning. In the early stages, the process of structure learning heavily depended on the utilization of expert information for the direct construction of Bayesian networks. As the quantity of samples increases, the initial approach becomes obsolete [99]. The fundamental approach involves employing a scoring function to assess the efficacy of the Bayesian network structure and thereafter choosing the structure with the highest score [100]. The establishment of the BBN structure is achieved through the processes of structure learning and structure optimization. Subsequently, the BBN undergoes parameter learning to compute the conditional probability of each node variable. The present work employs the EM algorithm for parameter learning. The EM method comprises two primary stages, namely the expectation step and the maximization step [101]. The present study uses the GeNIe3.0 software to establish correspondence between the variables present in the sample data and the nodes inside the Bayesian Belief Network (BBN). The parameter learning of the BBN can be accomplished using the EM technique. The Conditional Probability Tables (CPTs) of all nodes are acquired in this manner, serving as the foundation for the development of the DBN and inference of the model.

In this study, the consistency between the logical relationships and conditional probabilities of the nodes in the model and the real scenario is assessed through the use of 10 test samples as data for model validation. The subsequent paragraphs outline the procedural procedures involved in the validation of the model. The procedure involves the following steps: (1) incorporating the test samples as evidential data into the constructed Bayesian Belief Network (BBN) model; (2) utilizing the inference function of the BBN model to derive the probability distribution of decision-making risk; (3) comparing the predicted state with the fundamental level to assess the validation results; (4) repeating the iterative process outlined above to independently validate the remaining nine test samples; and (5) ultimately obtaining the validation results for all ten samples. A model is deemed valid when the proportion of measurement outcomes to the whole number of test data surpasses 80% [102].

3.3. Model Reasoning

There are four types of model reasoning, including sensitivity analysis, impact chain analysis, predictive analysis, and diagnostic analysis. First and foremost, sensitivity analysis refers to the process of identifying influential elements among several uncertainties that exert a substantial influence on the risk associated with a choice [103]. This analysis can ascertain the primary determinant of decision-making risk when risk components undergo alterations. Additionally, influence chain analysis is a method that characterizes the level of interaction among nodes and illustrates the interdependency of conditional probabilities between these nodes. The objective is to investigate the longest sequence of causal influences that lead to a particular outcome. The magnitude of the effect between the parent nodes and the offspring nodes is characterized by the width of a directional arc. Thirdly, predictive analysis utilizes the provided evidence to forecast the degrees of risk associated with decision making. The process of disseminating evidence pertaining to risk factors for decision making enables the revision of the probability distribution of risk levels within a Bayesian network, in light of newly acquired evidence. In addition, the diagnostic analysis uses the backward inference technique in Bayesian reasoning to identify the primary factor responsible for decision-making risk. The present study is capable of computing the posterior probability associated with each risk element in decision making. Furthermore, it can identify the reasons that are most suspicious in terms of risk reduction, thereby pinpointing the components that exert the greatest influence on decision-making risk.

4. Model Development

4.1. Factor Preparation

This study primarily employs direct routes, including in-depth interviews, to gather firsthand information from the research participants. Subsequently, it incorporates the identified risk variables [104,105,106]. Based on empirical evidence from engineering practice, the study selects 10 decisionmakers involved in the preproject stage, design stage, and construction stage as the participants for the interviews. The questionnaire (as shown in Appendix A) was distributed from April to June 2022, using a combination of offline and online methods. A total of 195 questionnaires were returned. The participants of the interview are affiliated with various academic institutions, such as universities and research institutes, as well as professional entities, including construction units, design firms, and consulting agencies, among others. The data shown in

Table 2. Descriptive statistical analysis table of respondents.

Sample Category	Variant	Sample Size	Percentage (%)
Sex	Male	175	94.1
	Female	11	5.1
Age	20~30 years	45	24.2
	31~40 years	60	32.3
	41~50 years	64	34.4
	>50 years	17	9.1
Educational background	Bachelor	136	73.1
	Master	26	14.0
	Doctoral	4	2.2
	Others	20	10.8
Working experience	Less than 5 years	117	49.16
	6~10 years	43	18.30
	More than 10 years	75	31.91
Project position	Senior management	35	18.8%
	Middle management	97	52.2%
	Below middle level	50	26.9%
	Others	4	2.1%

provide a comprehensive overview of the respondents’ basic information in the sample, encompassing variables such as gender, age, educational background, and years of working experience. This table effectively captures the diverse and distinctive backgrounds of the respondents. The content of the interview questionnaire is organized in a manner that allows for the ranking of decision-making risk variables in the model based on their likelihood and degree of influence on the target node. The study categorizes the subjects into five distinct levels, namely insignificant impact, minor impact, medium impact, major impact, and severe impact, in a sequential manner. Once the definition of the degree of leverage of the network nodes has been established, the nodes and their corresponding probability statistics are merged with the probability table of decision-making risk factors. This integration is then combined with the relevant data to determine the initial probability of each node within the Bayesian network, as depicted in

Table 3. The probability of integration of factors influencing risk.

Risk Node	Statistical Results (%)
	Negligible Impact	Minor Impact	Medium Impact	Large Impact	Severe Impact
Product/Service prices A01	20.1	12.7	27.52	24.34	15.34
Managerial operational level A02	12.7	21.17	28.06	31.21	6.86
Market demand A03	10.56	12.72	42.38	30.68	3.66
Integration of the bidding procedures of the two parties into one A04	14.28	26.99	26.99	24.34	7.4
Financing costs A05	14.27	20.64	35.51	24.84	4.74
Bidding competition A06	20.62	24.38	37.08	15.31	2.61
Contract negotiation A07	22.22	20.13	33.88	21.14	2.63
Completeness of contract documents A08	47.99	12.46	27.85	5.85	5.85
Quality of design solution A09	36.34	34.84	24.83	2.96	1.03
Access to finance A10	53.62	32.34	11.06	2.55	0.43
Operating cost A11	17.95	17.95	36.86	20.11	7.13
Adaptation of laws and regulations A12	16.23	14.15	37.77	27.86	3.99
Governance of the relationship between the social capitalist and the government A13	20.07	15.35	26.61	26.49	11.48
Project adaptation A14	6.71	7.54	16.93	54.74	14.08
Selection of consortium members A15	16.35	25.21	43.2	12.96	2.28
Multiple identities of social capitalists A16	33.56	24.18	25.91	12.47	3.88
Decision-making interventions by governments A17	26.25	30.05	20.61	17.76	5.33
Control structure A18	22.9	22.82	27.72	23.01	3.55
Stakeholder needs A19	20.92	25.79	36.46	15.87	0.96
Project schedule delay A20	62.8	25.92	7.96	2.64	0.68
Construction cost overrun A21	36.81	28.02	27.63	6.55	0.99
Engineering quality issues A22	24.86	13.22	36.68	22.65	2.59

.

4.2. BBN Model Design

The Bayesian network structure model plays a crucial role in Bayesian network inference analysis, since the precision of the network model directly impacts the dependability of the outcomes obtained by Bayesian inference analysis. The 184 training samples are imported into the GeNIe3.0 software and afterward configured with the scoring-search method. The scoring function employed in this study is the minimum description length (MDL) scoring [107], whereas the search strategy utilized is the greedy algorithm. Subsequently, the established causal connections among nodes are incorporated into the software as prior knowledge, following which the BBN model can be executed by machine learning techniques. Nevertheless, it is important to note that there may exist irrational causal connections among nodes in the first framework, such as factors A20 and A21. Therefore, this study utilizes expert knowledge to assess and enhance the causal linkages between variables in the preliminary design. The 10 interviewed experts analyzed and discussed the preliminary BBN model. They conclude that the connection between the A20 and A21 factors is in line with engineering principles and should be preserved. This step is crucial to further refine and optimize the network structure. The present study involves the modification and enhancement of the logical connections among nodes in the initial design through the integration of expert viewpoints. The network structure was reoptimized, and the structure of the BBN network is depicted in Figure 1.

Prior to advancing to future investigation, it is important to conduct model validation for the created BBN model pertaining to decision-making risk in mega infrastructure projects. The probability of all nodes is aggregated from the prior research. In this phase, model validation is conducted using a set of 10 test samples. These samples are imported into the created BBN model to validate the model. The purpose of model validation is to assess the consistency between the logical relationships and conditional probabilities of the nodes in the model and the real-world context of decision-making risk in mega infrastructure projects [108]. This process allows for an evaluation of the accuracy and validity of the model under specific variable conditions. The model validation findings are depicted in Figure 2.

Based on optimum Bayesian decision theory [109], the decision state that is considered final is the one with the highest probability in the probability distribution. In other words, the prediction state corresponds to the state in the probability distribution with the highest value. According to the data presented in Figure 2, the probability distribution of the decision-making risk associated with sample P1 is as follows: negligible (3%), minor (92%), medium (2%), big (2%), and severe (2%). According to the model inference, the anticipated condition of this sample is of lesser significance. The current level of risk associated with decision making is quite low in contrast. Consequently, the projected outcome of this particular sample aligns with the actual outcome, indicating a high degree of precision in the measurements. Similarly, the projected states of the remaining nine test samples are compared with the corresponding actual values. This provides evidence for the soundness of the proposed model. Hence, the BBN model proposed in this work demonstrates its feasibility and potential for utilization in model analysis and subsequent research endeavors.

4.3. DBN Model Building

The DBN is a temporal extension of the BBN, serving as a dynamic model that exhibits changes over time [110]. A DBN is a graphical model that portrays a system by capturing its state at different points in time, from the initial time to the final time. Every individual snapshot corresponds to a self-contained Bayesian network that accurately represents the system’s state at a specific moment in time. To depict the propagation relationship of node states at different points in time, a causal link is established between the pertinent nodes in the preceding and subsequent snapshots. The utilization of the DBN model enables the acquisition of posterior probabilities pertaining to decision-making risk throughout a certain period, hence enhancing our comprehension of the underlying dynamics associated with decision-making risk. The expansion of the BBN to the DBN can be achieved through the utilization of the Transfer Probability Table (TPT), which is a table that represents the probabilities of transitioning between states. The present study defines the TPT for decision-making risk between moment t − 1 and moment t, utilizing expert knowledge. The findings of this analysis are presented in

Table 4. TPT for decision-making risk nodes.

Decision-Making Risk Node (t − 1)	Decision-Making Risk Node (t)
	Negligible (t)	Minor (t)	Medium (t)	Large (t)	Severe (t)
Negligible (t − 1)	0.06	0.02	0.10	0.07	0.03
Minor (t − 1)	0.02	0.50	0.15	0.10	0.05
Medium (t − 1)	0.05	0.15	0.65	0.10	0.05
Large (t − 1)	0.04	0.08	0.13	0.60	0.15
Severe (t − 1)	0.02	0.03	0.15	0.20	0.60

.

Based on the decision-making risk nodes presented in Table 4, when the decision-making risk is classified as severe at time t − 1, there is a 0.60 probability that it will remain in the severe state at time t. Additionally, there is a 0.20 likelihood that it will transition to the large state, a 0.15 likelihood that it will transition to the medium state, a 0.03 likelihood that it will transition to the minor state, and a 0.02 likelihood that it will transition to the negligible state. In a similar vein, drawing from the practical expertise in project management, the TPT pertaining to decision-making risk can be derived for alternative states by considering the period from time t − 1 to time t. Following the establishment of the preliminary network model and the definition of the TPT, the construction of the DBN model is undertaken, as depicted in Figure 3.

5. Analysis and Results

5.1. Sensitivity Analysis

Sensitivity analysis is a method used to elucidate the relationship between decision-making risk factors and the extent to which changes in these factors impact them. This is achieved by systematically altering the values of pertinent variables, one at a time, to discern patterns and understand the effects. The primary objective of this study is to conduct a sensitivity analysis on a decision-making risk BBN model. The purpose of this analysis is to ascertain the relative importance of different risk factors in influencing decision-making risk as they undergo variations. The aforementioned risk factors have the potential to undergo modest modifications, hence resulting in substantial alterations in the likelihood of decision-making risk manifestation. This approach facilitates the concentration of project decisionmakers on risk-influencing factors that impact decision making, hence enhancing management efficiency. Once the target node is established in this study, simulations are conducted utilizing the GeNIe3.0 software. Subsequently, the impact on the posterior probability of the target node is computed. The nodes representing risk factors are assigned specific colors to visually reflect their sensitivity. The intensity of coloration indicates the level of influence that risk variables have on decision making, with darker colors representing a higher degree of influence. The outcomes of the sensitivity analysis are depicted in Figure 4.

As seen in Figure 4, sensitive factors affecting project completion include integration of the bidding procedures of the two parties into one (A04), operating cost (A11), stakeholder needs (A19), project schedule delay (A20), construction cost overrun (A21), financing costs (A05), access to finance (A10), governance of the relationship between the social capitalist and the government (A13), and control structure (A18). Sensitive factors can be ranked according to the exact sensitivity value of the variables as follows: integration of the bidding procedures of the two parties into one (A04) > stakeholder needs (A19) > construction cost overrun (A21) > operating costs (A11) > project schedule delay (A20) > control structure (A18) > financing costs (A05) > governance of the relationship between the social capitalist and the government (A13) > access to finance (A10). Minor alterations in the aforementioned risk factors can exert a substantial influence on the likelihood of decision-making outcomes. Hence, it is imperative to give due consideration to these delicate elements and implement suitable steps to mitigate the likelihood of risks associated with decision making.

5.2. Influence Chain Analysis

The influence chain analysis is employed to delineate the extent of interaction between nodes. The concept of influence chain analysis pertains to the assessment of interdependencies among conditional probabilities to identify the most probable pathway leading to a particular outcome. The breadth of the link is indicative of the strength of the relationship between the variables of the interconnected nodes, specifically denoting the level of impact exerted by the parent node on its offspring nodes. A link can be considered a causal chain that possesses optimal effectiveness when it comprises many nodes that exhibit robust impact linkages, and the desired target node is encompassed within such a link. This study examines the effect chain analysis of the DBN model, and the findings are presented in Figure 5.

As seen in Figure 5, several chains of influence emerge, as shown in the thickened chains. The first one is “operating costs (A11) → access to finance (A10) → financing costs (A05) → integration of the bidding procedures of the two parties into one (A04) → decision-making risk (DR)”. The second one is “operating costs (A11) → access to finance (A10) → bidding competition (A06) → contract negotiation (A07) → multiple identities of social capitalists (A16) → stakeholder needs (A19) →decision-making risk (DR)”. The third one is “operating costs (A11) → access to finance (A10) → bidding competition (A06) → contract negotiation (A07) → multiple identities of social capitalists (A16) → stakeholder needs (A19) → governance of the relationship between the social capitalist and the government (A13) → project adaptation (A14) → decision-making risk (DR) ”. The fourth one is “construction cost overrun (A21) → project schedule delay (A20) → stakeholder needs (A19) → decision-making risk (DR)”. The fifth one is “construction cost overrun (A21) → project schedule delay (A20) → control structure (A18) → project adaptation (A14) → decision-making risk (DR)”.

The inclusion of various impact factor nodes is a result of the presence of multiple influence chains. The direct derivation of the few major influence factor nodes that exert the most significant impact on decision-making risk is not feasible. Additional predictive and diagnostic analyses should be conducted to arrive at a scientific conclusion for the study through simulation analysis.

5.3. Predictive Analysis

Predictive analysis utilizes the provided evidence to forecast the degrees of risk associated with decision making. The process of disseminating evidence pertaining to risk variables in decision making enables the revision of the probability distribution of project decision-making risk levels within a Bayesian network in light of newly acquired evidence. The severity of various decision-making risk factors is indicated as evidence, and the software GeNIe3.0 is employed to forecast the likelihood of project decision-making risk across multiple scenarios. Figure 6 depicts the probability of encountering project decision-making risks at varying levels when conducting predictive analyses in different scenarios.

From Figure 6, it can be seen that from a short-term perspective (Time = 0), the probability of the decision-making risk level being “severe” is the highest when the A19 (stakeholder needs) status is set to “severe”, and the probability of the decision-making risk level being “severe” is the second highest when the A04 (integration of the bidding procedures of the two parties into one) status is set to “severe”; meanwhile, it is evident from the figure that P(DR = severe∣A19 = severe) > P(DR = severe∣A04 = severe) > P(DR = severe∣A14 = severe) > P (DR = severe∣A13 = severe) > P(DR = severe∣A15 = severe) > P(DR = severe∣A18 = severe) >P (DR = severe∣ A05 = severe) >P (DR = severe∣A20 = severe) > P (DR = severe∣A16 = severe) > P(DR = severe∣A21 = severe) > P (DR = severe∣A07 = severe) > P (DR = severe∣A06 = severe) > P (DR = severe∣A10 = severe) > P (DR = severe∣A11 = severe), indicating that stakeholder needs (A19), integration of the bidding procedures of the two parties into one (A04), project adaptation (A14), and governance of the relationship between the social capitalist and the government (A13) are the critical influencing factors in decision-making risk.

The risk likelihood associated with decision making in distinct states exhibits temporal variation from a long-term perspective, namely after Time = 1. The decision-making risk probability reaches its highest value and then stabilizes at the middle stage. The observation reveals that when many critical factor nodes are assigned the condition of “severe”, the risk curve associated with the “severe” state tends to exhibit an initial increase, followed by a period of stabilization. Nevertheless, the likelihood ultimately stays quite modest. The risk curve for both large and medium states exhibits a gradual increase, followed by a period of stabilization. The primary rationale behind the use of DBN is its ability to retain and aggregate the outcomes of previous instances of reasoning. This implies that the risks encountered at the beginning stages continue to impact decision-making risk over an extended period. This highlights the imperative of managing risk variables from an alternative perspective. Over some time, the network model can integrate information from diverse elements to generate accurate predictions regarding the target nodes in the long term. The dynamic character of decision-making risk might become evident over time in this manner. To summarize, the main focuses to enhance the risk of decision making in mega infrastructure projects should be stakeholder needs (A19), integration of the bidding procedures of the two parties into one (A04), adaptation to laws and regulations (A12), governance of the relationship between the social capitalist and the government (A13), selection of consortium members (A15), and control structure (A18).

5.4. Diagnostic Analysis

This diagnostic analysis employs the backward inference technique within the framework of Bayesian reasoning to ascertain the primary factors that are most likely to have a substantial influence on the level of risk associated with decision making. The study has the capability to compute the posterior probability associated with each risk factor involved in the decision-making process. The identification of the most suspicious reasons is facilitated by the reduction in decision-making risk. Therefore, the components inside the Bayesian network that exert the most substantial influence on decision-making risk are determined. The decision-making risk node’s state is designated as “severe”, and the GeNIe3.0 software is utilized to compute the posterior probability of the nodes associated with the key influencing elements of the decision-making risk. The posterior probability of the nodes in the diagnostic analysis for various situations is presented in Figure 7.

As can be seen from Figure 7, the posterior probability of project adaptation (A14) has a maximum value at T = 0 when the node status is “severe”. Similarly, the posterior probability of project adaptation (A14) is also maximum for the node state “severe” at T = 30. According to the probability distribution of the diagnostic analysis of different nodes, when P (DR = severe) = 1, the size of the posterior probability of the node is A14 > A19 > A10 > A04 > A11 > A05 > A16 > A20 > A18 > A07 > A06 > A21 > A13 > A15. Therefore, project adaptation (A14), stakeholder needs (A19), access to finance (A10), integration of the bidding procedures of the two parties into one (A04), operating costs (A11), financing costs (A05), and multiple identities of social capitalists (A16) are the most influential factors leading to reduced decision-making risk. Put differently, a heightened level of one or more of these characteristics will exert a significant influence on the level of risk associated with decision making. Hence, it is imperative to prioritize the management of risk elements to enhance the decision-making process pertaining to megaprojects.

6. Discussion

6.1. Discussions of Findings

The above simulation analysis of the decision-making risk model using GeNIe3.0 can be summarized as follows. Several factor nodes according to the sensitivity analysis results are A04, A19, A21, A11, A20, A18, A05, A13, and A10. Several factor nodes according to the predictive analysis are A19, A04, A14, A13, A15, A18, and A05. According to the magnitude of diagnostic analysis, several key nodes are derived as A14, A19, A10, A04, A11, A05, and A16. Combined with several critical causal chains derived from the influence chain analysis, it can be concluded that integration of the bidding procedures of the two parties into one (A04), financing costs (A05), project adaptation (A14), control structure (A18), and stakeholder needs (A19) are the most critical risk factors affecting decision-making risk.

The results of this study are compared and analyzed with the results of existing studies. Similarly, Liu et al. [111] used the same research method to assess the risk factors in five dimensions of urban rail transit PPP projects. Overseas, some scholars also identify the risks from different dimensions, which form a comparative analysis with this study. Nabawy et al. [112] developed a computerized risk identification framework that identified the challenges faced in the construction of megaprojects, including three dimensions of influencing factors, including scarcity of construction resources, mismanagement of projects, and financial funding issues. Ullah et al. [113] identified five dimensions of risk factors affecting risk appetite based on their literature study, including organizational, decisionmaker, environmental, and project factors. Kimiagari et al. [114] used the fuzzy Delphi method to identify three basic risk categories for mega- and complex engineering and construction projects, including business, legal, and operational risks.

It is important to note that the occurrence of project schedule delays will lead to cost overrun [115,116,117,118]. Lee analyzed 161 projects and found that construction delays were the main cause of cost overruns [119]. Kaliba et al. found that project delays led to period cost overruns, including material purchases, staffing issues, specification changes, and labor disputes [120]. In their study, Stewart et al. explored the risk of different construction delay drivers and the implications on total installed cost [121]. Natawidjana and Nurasiyah used descriptive analysis to determine that an acceleration in the duration of each activity would lead to an increase in labor costs [122]. The aforementioned factors are likely to lead to an increase in expenses during the implementation of the construction endeavor.

6.2. Model Application

The Yiwu Mall Avenue project is chosen as a case study to assess the alignment between the decision-making risk model of mega infrastructure projects under the EPC + PPP model and a real-world scenario. The decision-making risk of the Yiwu Mall Avenue project is processed and analyzed using the inference function of GeNIe3.0 software. The DBN model of the Yiwu Mall Avenue project is depicted in Figure 8. The PPP + EPC model of the Yiwu Mall Avenue project incorporates an examination of its actual construction, thereby considering the interests of investment, construction, and social impact. The government utilizes the PPP model to facilitate the integration of social capital. It is the government’s duty to effectively engage social capitalists, enhancing the social benefit threshold while maximizing overall comprehensive advantages. Contractors involved in the construction process should take into account the standard construction procedures for the project and also consider the viewpoint of their investors in terms of project quality and timely arrangements. Simultaneously, the establishment of this initiative as an exemplar for other collaborative endeavors, together with its consequential societal advantages, has considerable importance.

6.3. Risk Governance Strategies

This section focuses on the governance strategy for the five key risk factors obtained above that affect decision-making risk. The “two-tenders-and-one-tender” bidding method can enhance the integration of the four phases of financing, design, construction, and operation of megaconstruction projects. It is recommended that project stakeholders, particularly the government and social capital entities, establish a dedicated team with expertise in contract review. Meanwhile, to mitigate the potential financial constraints faced by the project company, it is imperative for the company to proactively assess the risks associated with project financing [123], and it is imperative to conduct thorough monitoring and supervision of critical components that have the potential to contribute to such overruns [124]. Under this model, there may be instances where the government is involved in the formulation or revision of relevant policies. To address this issue, social capitalists need to engage in regular information exchanges on the progress of the project and related risks to establish a stable partnership with the government. Furthermore, megaengineering projects implemented using the EPC + PPP model have the distinct feature of including several stakeholders [125]. Within this intricate environment, it is imperative to issue well-defined documents that delineate the rights and duties of both the government and the social capital party within the project’s parameters. Eventually, it is imperative for stakeholders to build a robust communication and consultation process as a fundamental approach to addressing and resolving divergences [126].

7. Conclusions and Future Works

This study takes the risk management of megaprojects under the EPC + PPP model as the research object and launches an in-depth study on the identification, analysis, and response to the risk of megaprojects under the EPC + PPP model through quantitative analysis methods. Through literature combing, 22 influencing factors are identified. Among them, A04\A05\A14\A18\A19 are identified as the most critical risk factors affecting decision-making risk under the EPC + PPP model in China. Then, the BBN model and DBN model are constructed based on machine learning and expert knowledge. Dynamic simulation analyses are carried out by GeNIe3.0 software, and corresponding risk management strategies for megaprojects are proposed.

This report serves as a crucial resource for decisionmakers involved in the management of decision-making risks in megaprojects. The initial step involves the construction of a DBN model that encompasses the influential factors affecting risk in megaproject decision making. This model is grounded in Bayesian network theory and incorporates the temporal dimension, enabling the integration of project risk management theory and practice. Consequently, this advancement contributes to the enhancement of project risk management theory. Furthermore, there is a paucity of research concerning the hazards associated with decision making in the context of megaprojects. A comprehensive examination and investigation of the factors that influence decision-making risks in megaprojects under the EPC + PPP model has not been conducted systematically. This study uses Bayesian network analysis and empirical analytic techniques to perform quantitative analysis.

The method created has practical implications. This study aims to investigate risk-coping mechanisms while taking into account various stakeholders. This approach mitigates the potential for disputes arising from the adoption of risk-coping measures that cater solely to the needs and preferences of one party, ensuring a more harmonious outcome. Furthermore, this study chose the Yiwu Mall Avenue project as a means of validating the case. The findings indicate that the degree of influence of decision-making risk aligns with the current circumstances of the project. In the context of having adequate data, the use of DBN enables the prediction of risk levels at various time intervals. Furthermore, it allows for the identification of the changing trends of these risk levels over time. This information serves as a valuable reference point for managers in making informed decisions regarding risk governance.

Nevertheless, the constructed DBN model exhibits several limitations. The construction of logical relationships in the models of this study is achieved using machine learning techniques. Certain results will undergo human correction to some extent, guided by expert judgment, which introduces a subjective element. The limits of the conclusions can be further elucidated and enhanced in further studies. It is important to note that the network configuration of decision-making risks and the conditional probabilities among the elements might exhibit substantial variations over time and throughout different stages of a project. Hence, it is imperative for future research endeavors to consider the distinct attributes of decision-making risks throughout various stages of a project. Furthermore, the constructed DBN model exhibits a certain degree of application. The primary focus of this research revolves around the examination of project risks within the transportation sector. Other types of projects, such as those related to energy and water, are taken into account. This research will undergo further refinement and continuation in subsequent investigations. Future research endeavors may delve into the convergence of technical advancements and the management of potential risks within large-scale infrastructure initiatives in China.