Integrated Fuzzy Technique for Order Preference by Similarity to Ideal Solution and Emotional Artificial Neural Network Model for Comprehensive Risk Prioritization in Green Construction Projects

Abstract

With the rapid growth of green construction projects (GCPs) in Saudi Arabia, managing the associated risks has become crucial to ensuring project success and sustainability. These projects face a range of challenges, including socio-economic, environmental, and technical risks that need to be carefully identified and prioritized. This study systematically identifies and prioritizes the risks in GCP using a hybrid model combining fuzzy TOPSIS and an Emotional Artificial Neural Network (EANN). The focus of this study is on the risk management of the green construction industry in Saudi Arabia. Based on expert evaluations, low-quality materials and equipment (Likert scale mean is 4.71) and stakeholder resistance to adopting green ideas (4.67) emerged as the most critical risks. The fuzzy TOPSIS analysis assigned the highest weight to risk probability (0.174), followed by outcome (0.137), and vulnerability (0.123). The EANN refined the risk rankings, confirming the importance of these risks. The findings suggest that risk management strategies should prioritize material quality and stakeholder engagement, while environmental risks, ranked lower, are less of a concern. This hybrid model provides a robust framework for effective risk management, with practical implications for enhancing the sustainability and success of GCP.

1. Introduction

The construction industry, especially in rapidly developing regions, plays a pivotal role in shaping the environmental and societal landscape [1,2,3,4]. As urbanization intensifies, the need for sustainable building practices has become more pronounced to mitigate the negative environmental impacts [5,6,7]. GCPs are seen as a viable solution to minimize the environmental risks associated with traditional construction methods. These projects aim to reduce the carbon footprint, enhance energy efficiency, and utilize sustainable materials. However, despite their potential benefits, GCPs are not without their risks. Identifying and managing these risks is crucial to ensuring the long-term success and sustainability of such initiatives [8,9].

Risk management plays a pivotal role in GCP management, exerting a substantial influence on the overall success of projects [10,11]. The scope of uncertainty in project environments is extensive, encompassing various factors such as ambiguities in project design and procurement processes, uncertainties in foundational assumptions and initial estimates, and potential misalignments in project objectives [12,13]. These inherent uncertainties emphasize the vital role of risk management in ensuring effective project planning and execution. This underscores the significance of risk management in mitigating potential risks and successfully achieving project objectives. By proactively addressing uncertainties, stakeholders can make informed decisions that enhance the likelihood of project success and reduce the impact of unforeseen challenges. Risk management enhances the effectiveness and efficiency of projects by proactively identifying uncertainties before they escalate into crises and by balancing potential threats and opportunities. It encompasses critical processes such as identifying project risks, evaluating and prioritizing them based on their significance, accounting for the interdependencies among uncertainties, and formulating appropriate risk response strategies. These processes demand a comprehensive and systematic analysis of the project, ensuring that the risks are managed in a way that aligns with the overall project objectives and resources [14,15].

Given the inherent complexities of GCPs, it is essential for the involved organizations and corporations to utilize specialized project management tools, techniques, and processes, including robust risk management strategies [16]. By effectively identifying and prioritizing risks, these entities can significantly contribute to the successful execution of green construction initiatives, ensuring that potential challenges are addressed proactively to enhance project outcomes [17]. One of the main challenges in green construction is the inherent uncertainty and complexity of the risks involved. Environmental risks, coupled with socio-economic and political factors, contribute to the uncertainty of project outcomes. Risks such as low-quality materials, stakeholder resistance, and unrealistic project goals can severely impact the efficiency and accuracy of the project’s objectives [18]. For green building projects to be successful, a robust risk management framework is required, one that not only identifies risks, but also prioritizes them based on their potential impact. This becomes particularly important in developing economies like Saudi Arabia, where the green building concept is still gaining traction, and project execution often faces resistance from stakeholders unfamiliar with its benefits [19,20,21,22].

Paul and Taylor [23] demonstrated that green buildings have higher indoor environmental quality than conventional buildings, which increases resident satisfaction and labor efficiency. According to Castleton et al. [24], green roofs provide passive cooling by blocking sunlight from reaching the building infrastructure. The authors demonstrated that, despite poor insulation, older buildings require high levels of insulation to fully benefit from green roofs under the current building regulations, which are heavily influenced by the building’s annual energy consumption. Perez et al. [25] proposed a classification of green vertical systems for buildings. The goal of this classification is to learn how to distinguish between systems. In addition, the authors demonstrated that green facade mechanisms could function as passive energy storage systems. Isnin et al. [26] explored the challenges of using unfamiliar materials in adaptive green building projects. Their study highlights the adverse effects of building materials on sustainability and global concerns about toxic content information. Discussions on green construction reforms complicate building adaptation and increase capital availability. To improve green building practices, better management of materials is needed during construction, occupancy, and operation. GhaffarianHoseini et al. [27] examined sustainable energy performance in green buildings, aiming to clarify sustainability in this context. They focus on identifying the key parameters that contribute to effective energy performance in green buildings. The study analyzes current trends and their significant impacts on sustainability, noting that the sustainable energy performance of green buildings is crucial for reducing carbon dioxide emissions and energy consumption. Additionally, the review highlights the ongoing challenges and emphasizes the need for energy-efficient solutions to guide future energy demands. The findings underscore the importance of integrating renewable energy systems and addressing challenges related to cost, maintenance, and operation.

Several studies have explored risk management in construction projects, primarily using traditional methods like AHP, TOPSIS, and fuzzy logic. Zhao et al. [28] used a fuzzy hybrid approach to assess the risk of one case of green projects in Singapore. The proposed model was used to calculate the probability and impact of risks. The findings revealed that the “improper cost estimate” factor ranked highest in this evaluation. Yang and Zou [29] introduced a social network algorithm model designed to assess stakeholders and the associated risks in green building projects. Their model consists of interconnected stakeholders and risks, with the dependencies between them being represented as nodes. As the number of nodes increases, the complexity of the network grows. The authors employed a multi-criteria decision-making approach to evaluate the relationships between risks and stakeholders. The authors identified the specific risks associated with each stakeholder, emphasizing the importance of understanding these connections in managing risks effectively within green building initiatives. Zhang [30] used a research station project to solve the problem of selecting a risk response strategy while taking into account risk interdependence. The author demonstrated that paying little or no attention to the risk of interdependence reduces the expected utility and raises implementation costs. These studies emphasize the importance of multi-criteria analysis but often lack adaptability to dynamic risk profiles. More recent research has integrated machine learning models, yet these are generally static and do not provide real-time adaptability to changing project conditions.

Table 1. Literature summary regarding risk assessment in construction projects.

Study	Methodology	Key Findings and Gaps
[31]	Review of MCDM in construction project	Demonstrated MCDM’s effectiveness in structured risk assessment limited exploration of machine learning for adaptive risk ranking
[32]	Literature review alongside questionnaire design and mathematical analysis	Identified risks in GCPs, without consideration of changing environmental regulations, evolving technology, and material availability factor
[33]	Building economic risk assessment with photovoltaic technology applying the Bivariate Probit model	Identified obstacles of adoption of photovoltaic systems in the residential sector in an urban environment without consideration of the dynamic substance of fluctuating regulatory environmental factors
[34]	Risk evaluation in residential sector with solar energy utilization by DEMATEL-ANP	Evaluated risk for building-integrated photovoltaic projects, which lack adaptability to the dynamic, evolving risks inherent in green construction
[35]	Safety management in construction project using ANP	Identified and prioritized potential risks in the construction sector with the use of static models, which lacked adaptability for the dynamic and complex nature of green construction projects
[36]	Risk assessment at the planning stage by expert opinion using the Delphi technique and ANN	Cost risk estimation in residential projects primarily focuses on static models that assess risks at the early planning stages but lack the capacity to adapt to dynamic changes throughout the project lifecycle. There is also a lack of social risk assessment
[37]	Risk analysis in residential real estate investments utilize fuzzy cognitive mapping	Understanding static cause-and-effect relationships between risk determinants in the residential sector. However, the study did not offer a vast portfolio of techniques
[38]	NVivo tool for qualitative risk analysis with interview data collection	Identified common risks in mega projects. However, there is a lack of adaptive capabilities to address the evolving and complex nature of risk in green construction projects for effective decision making
[39]	Delphi survey, alongside an extensive literature review and TOPSIS method	Identified 82 potential risks in complex building projects and highlighted external factors (e.g., international relations) affecting construction risks without real-time adaptability
[40]	Provided static frameworks focused on fixed dimensions of risk DEMATEL and matter-element extension models	Construction risk assessment in a building-integrated photovoltaic project with a lack of flexibility to adapt to the dynamic and evolving nature of GCPs

shows a summary of the literature regarding risk factor evaluation in construction projects.

The current risk management approaches in green construction are often fragmented, focusing on either technical or financial aspects without providing a holistic view of the challenges faced. While traditional risk assessment models offer a structured way to rank risks, they may fall short in addressing the dynamic and uncertain nature of green construction risks. The integration of machine learning techniques with fuzzy logic provides an innovative approach to tackle this gap. Machine learning can help to refine risk assessment by predicting potential risk factors and their outcomes, while fuzzy logic allows for the handling of uncertainties inherent in human judgment, especially when assessing environmental and social criteria. Hwang et al. [32] performed a risk assessment for green residential construction projects in Singapore, focusing on identifying and evaluating project-specific risks, comparing them to traditional construction, and exploring mitigation strategies. The study found that the five most significant risks in green residential projects are as follows: “complex approval processes”, “high upfront costs”, “unclear requirements from owners”, “shortage of skilled labor”, and “lack of green materials and equipment”. Polat et al. [41] identified the risks related to materials used by contractors in the construction phase of green building projects. They recognized 25 risk factors and developed a questionnaire, which was distributed to a targeted group of contractors, designers, and consultants. The responses allowed them to rank and assess the impact of these risks on the costs and timelines of green projects. Ultimately, the authors concluded that “neglect of adaptation in green projects” is one of the most significant risk factors affecting both costs and schedules.

Wang et al. [42] examined the impact of artificial intelligence (AI) technologies on the construction sector, which is continuously evolving with the adoption of AI-driven advancements. Employing a hybrid multi-criteria decision-making framework that combines the Delphi method, analytic network process (ANP), and the fuzzy TOPSIS, the authors assessed the role of AI in construction. The ANP approach systematically determines the relative importance of various AI technologies, informed by expert insights from the Delphi survey, while fuzzy TOPSIS is applied to rank these technologies for optimal integration in construction. The TOPSIS results showed that robotics and automation are the most favorable AI tools, followed by building information modeling, while computer vision ranked lowest. However, the authors’ focus was on conventional buildings, and their evaluations were not related to green buildings. Nafei et al. [43] presented a decision-making framework, smart TOPSIS, designed to handle complex environments where uncertainty and indeterminacy play a significant role. By integrating TOPSIS with machine learning techniques. The proposed model utilizes a frequency-based ranking system in conjunction with neural-network-driven machine learning. The proposed framework creates a more adaptable and resilient multi-attribute decision-making solution. The results showed that the proposed model improves ranking precision and computational speed, making it applicable to a wide range of real-world decision-making scenarios beyond green supplier selection. Integration of TOPSIS and machine learning methods has attracted the attention of researchers due to increasing the quality and accuracy of the results. Nilashi et al. [44] presented a hybrid approach to analyze online reviews for sustainable green hotel development, focusing on factors influencing travelers’ choices. The method combines cluster analysis for textual review analysis, TOPSIS for feature ranking, and a Neuro-Fuzzy model to assess customer satisfaction. The findings offer valuable insights for both travelers making hotel choices and managers seeking to enhance service quality and marketing strategies. Jena and Pradhan [45] introduced an approach to earthquake risk assessment by integrating artificial neural network cross-validation with an AHP-TOPSIS hybrid model, tested in Aceh, Indonesia. The results showed that the adaptable model improves predictive accuracy over traditional methods, supporting targeted earthquake risk mitigation and planning across regions. Rafiei-Sardooi et al. [46] addressed urban flood risk in an Iranian city, using a hybrid model combining machine learning and TOPSIS for decision making.

Previous research indicates that traditional risk assessment methods in GCPs often focus on either technical or financial aspects in isolation, lacking a comprehensive view of the associated challenges. Furthermore, conventional models may fall short in addressing the dynamic and complex nature of GCPs. For instance, they struggle to adapt to rapid changes in technology, evolving environmental regulations, and shifting social priorities. Given the dynamic characteristics of green buildings, there is a critical need for advanced methodologies that can provide adaptive and holistic risk assessments, enhancing the precision and relevance of risk management strategies in this field. This study proposes a novel hybrid approach to risk assessment that combines the fuzzy TOPSIS method with advanced Emotional Artificial Neural Networks (EANNs). This integrated model focuses on assessing and prioritizing the risks associated with GCPs in Saudi Arabia. A comprehensive review of technical, economic, and environmental risks in GCPs has been conducted; moreover, by combining expert knowledge from the green construction field with sophisticated computational techniques, this approach provides a more precise and adaptable risk assessment framework. It allows stakeholders to identify and rank critical risks based on their importance to project success. Furthermore, this comprehensive framework is a valuable resource for construction industry decision makers, allowing them to effectively mitigate risks while promoting project sustainability.

2. Materials and Methods

Decision making involves clearly defining objectives, identifying potential solutions, evaluating their feasibility, analyzing the consequences and outcomes of each option, and ultimately selecting and implementing the best course of action. The quality of management is largely determined by the quality of these decisions, as they directly impact the effectiveness of plans and programs, the success of strategies, and the overall quality of the outcomes achieved [47]. When decisions are made based on multiple criteria, the decision maker is more likely to be pleased and satisfied. Criteria can be quantitative or qualitative [48]. Managers can make rational decisions by taking into account various decision-making criteria, which may conflict with one another [49]. The main purpose of this research is to identify and prioritize the risks of GCPs in Saudi Arabia. The development of green buildings faces several challenges, which are largely tied to environmental, economic, and socio-political factors. Despite the country’s commitment to sustainability, the high upfront costs of green technologies, a lack of awareness among stakeholders, and resistance to adopting new construction methods present significant barriers in Saudi Arabia. Additionally, the scarcity of locally sourced green materials and limited regulatory frameworks further complicate the widespread adoption of sustainable construction practices. Green building projects in Saudi Arabia encounter a range of risks that can significantly impact their success. One major concern is the fluctuation of currency exchange rates, which can affect the cost and availability of imported eco-friendly materials essential for sustainable construction. Additionally, there is often resistance from various stakeholders toward adopting sustainable innovations, which can hinder project implementation and acceptance. Moreover, uncertainty surrounding the long-term benefits of green building initiatives poses another challenge, as it may lead to hesitation among investors and decision makers. These factors underscore the critical need for robust risk management strategies tailored to the unique circumstances of GCPs. By effectively addressing these challenges, stakeholders can enhance the likelihood of successful project execution and contribute to the overall sustainability goals in the region.

This section provides a comprehensive overview of the methodology adopted for implementing the research, elaborating on the specific algorithms and processes employed.

2.1. Data Collection and Risk Identification

The participants ranged in age from 35 to 60 years, with an average age of 47 years, reflecting a seasoned group of professionals. Regarding educational qualifications, 65% of the experts held advanced degrees (Master’s or Ph.D.) in relevant fields, such as environmental science, civil engineering, and architecture, while the remaining 35% possessed bachelor’s degrees with extensive practical experience in the construction industry. The average length of work experience among the participants was 22 years, ensuring that the insights provided were grounded in practical, real-world experience.

Data collection involved a two-phase process. Initially, semi-structured interviews were conducted to gain qualitative insights into the experts’ perceptions of risks in green construction. These interviews helped to identify the key areas of concern, such as material quality, resistance from stakeholders to adopting green technologies, regulatory uncertainties, and economic volatility. Following the interviews, a comprehensive survey was developed based on the input received, which was then distributed to all participants for further quantitative assessment.

The survey included four detailed questionnaires with both open-ended and Likert-scale questions, focusing on the frequency, impact, and manageability of various risks. The experts were asked to rate the severity of each identified risk factor. Their responses were then compiled and analyzed, allowing for a thorough screening of risks. Only those risks that were rated as having a significant impact on project timelines, costs, or sustainability were included in the final list for further evaluation. This rigorous process ensured that the most relevant and critical risks were prioritized for analysis and risk mitigation in subsequent stages of the study. The experts completed a questionnaire on the importance of each risk in GCPs. Risks with values greater than the average of the total values were screened and confirmed. This is considered efficient, so it is chosen, and any risk with a value less than the average of the total values is deemed ineffective and removed.

The first questionnaire intended to identify the risks to GCPs. In this regard, the experts were first asked to rate the significance of the risks on a scale of 1 (very low importance) to 5 (very high importance). Then, the experts were asked to describe the risks in terms of specificity and generality. All components whose average degree of importance exceeded the overall average were chosen. The second questionnaire is used to prioritize the criteria in the TOPSIS technique. The experts were asked to rank the criteria based on their personal mentality. The final rating can be determined by taking into account the opinions of all experts, as well as the average rating of each criterion. The third questionnaire is distributed to experts after the criteria have been ranked in order to assess their relative importance. The fourth questionnaire, which is a matrix with risks and criteria, is given to the experts to evaluate each risk using a scale of 1 to 5.

2.2. Risk Assessment Criteria

After identifying the risks, a comprehensive set of assessment criteria was developed to evaluate their significance effectively. These criteria encompassed various factors, including the likelihood of occurrence, potential impacts on project timelines and costs, and the feasibility of mitigating these risks through established project management practices. To ensure a thorough evaluation, a multi-criteria decision-making framework was employed, which allowed for a holistic consideration of all relevant aspects of risk during the assessment process. Through a detailed review of the existing research literature on construction projects, a total of 25 risks associated with green building initiatives were identified and documented, as presented in

Table 2. The risk items in construction projects.

Risk Group	Risk Item	Reference
Construction (RG1)	Failure to comply with the standard in green construction (RG11)	[41,50]
	Delay in delivering the equipment to the location (RG12)	[51]
Economic and financial (RG2)	Additional costs due to green construction (RG21)	[32,52]
	The increase in time caused by green construction (RG22)	[50,52]
	Inflation (RG23)	[28,53]
	The effect of currency and interest rate fluctuations on the import of green materials (RG24)	[28]
	Incorrect forecasting of market demand (RG25)	[51,54]
Management (RG3)	Lack of management experience (RG31)	[29,30,50,54]
	Lack of knowledge about technology and green materials (RG32)	[29,54]
	Limited access to green suppliers (RG33)	[41,52]
	Lack of experienced construction companies in green construction (RG34)	[50,51]
	Lack of quantitative evaluation criteria and tools for green performance (RG35)	[50,52]
	Resistance from stakeholders to adopt green ideas (RG36)	[51]
Design and technical issues (RG4)	Failure to comply with green building factors (RG41)	[41]
	Lack of documents and information for novel green building technologies (RG42)	[51,54]
	Low-quality materials and equipment (RG43)	[54]
	Production limitations and new green building technology (RG44)	[54]
	Technical complexity (RG45)	[29,54]
Environmental (RG5)	Adverse weather and geological conditions (RG51)	[51]
Material (RG6)	Lack of clear definition of green building materials (RG61)	[41]
Human resource (RG7)	Damages caused by human error (RG71)	[32,51]
	Lack of skilled and experienced professionals (RG72)	[32]
Safety (RG8)	Accidents leading to disability or death (RG81)	[55]
	Accidents leading to injury (RG82)	[55]
	Lack of safety equipment (RG83)	[50]

. This systematic approach provides a clear foundation for analyzing the potential challenges and informing strategies for effective risk management in GCPs.

Meanwhile, after evaluating risk assessment criteria based on the literature, 12 criteria were identified to assess the risks of GCPs, as shown in

Table 3. Risk assessment criteria for GCPs.

	Criteria	Reference
1	Vulnerability	[53,56]
2	Threat	[57]
3	Outcome	[53]
4	Uncertainty	[57]
5	Reaction to risk	[53,58]
6	Prediction of risk	[59]
7	Risk management	[53,59]
8	Risk identification	[53,56,57,58]
9	Risk probability	[51,53]

.

2.3. Fuzzy TOPSIS and Criteria Weighting

In this section of this study, the fuzzy TOPSIS method was utilized to allocate weights and prioritize the identified risks. This methodology was chosen due to its effectiveness in managing the uncertainties and ambiguities that are commonly encountered in expert assessments. These uncertainties frequently arise from incomplete or vague information found in practical scenarios, such as risk evaluations in GCPs. By integrating fuzzy logic into the conventional TOPSIS framework, the approach facilitated more precise and flexible rankings of the risks involved, allowing for a more robust decision-making process. This enhancement provides decision makers with a clearer understanding of the potential risks, thereby improving their ability to implement effective risk management strategies [60,61,62].

TOPSIS is grounded in the idea that the best possible solution should simultaneously minimize its distance from the positive ideal solution (PIS) and maximize its distance from the negative ideal solution (NIS). In practice, this means identifying the alternative that is closest to the ideal scenario and furthest away from the worst-case scenario. When employing fuzzy TOPSIS, the process is enhanced by the use of fuzzy numbers, which serve to represent both the criteria weights and the performance ratings of each alternative option under consideration. Fuzzy numbers are particularly useful because they allow for a more comprehensive and flexible representation of uncertainty and imprecision, common in complex decision-making environments [63,64].

In risk assessments of GCPs, expert opinions often come with a degree of ambiguity, due to incomplete or uncertain data. Fuzzy TOPSIS addresses this issue by incorporating fuzzy sets that capture the vagueness in expert judgments, thereby providing a more detailed evaluation. This nuanced approach improves the accuracy of decision making by allowing for a richer, more adaptable comparison of risks. As a result, decision makers can achieve more reliable rankings of the alternatives and better manage the complexities involved in green building initiatives.

In the fuzzy logic approach, each risk criterion was evaluated using triangular fuzzy numbers (TFNs) to represent the degree of importance. A TFN is defined by three parameters, $\left(l,m,u\right)$, where $l$ is the lower bound, $m$ is the most likely value, and $u$ is the upper bound. These values were determined based on expert input, allowing for flexibility in the evaluation process. A TFN $\tilde{A}$ can be represented with Equation (1), as follows:

$$\tilde{A}=\left(l_A,m_A,u_A\right),$$

(1)

where $l_A$ represents the lowest value of the fuzzy number, $m_A$ represents the most probable value, and $u_A$ represents the highest value of the fuzzy number.

The fuzzy TOPSIS method involves several key steps to evaluate and prioritize risks in GCPs. First, the decision matrix undergoes fuzzification, converting crisp values into fuzzy numbers based on expert assessments. Next, fuzzy weights for each criterion are determined to reflect their relative importance. Following this, a normalized fuzzy decision matrix is constructed to ensure comparability across different criteria. The method then identifies the fuzzy positive ideal solution (FPIS) and fuzzy negative ideal solution (FNIS), representing the best and worst outcomes for each criterion, respectivley. The distances from both the FPIS and FNIS are calculated for each alternative, allowing for a performance evaluation relative to the ideal solutions. Finally, the closeness coefficient (CC) is computed to rank the alternatives based on their proximity to the FPIS, thereby facilitating informed decision making in the context of green construction initiatives.

The decision matrix was established, organizing the alternatives (risks) along the rows and the criteria along the columns. Each element within the decision matrix was subsequently converted into a fuzzy number, reflecting expert evaluations using terms such as “low”, “medium”, and “high”.

Each criterion was assigned a weight using fuzzy numbers, with expert opinions aggregated to calculate the fuzzy weight for each criterion. The aggregated weight for a criterion $C_i$ was represented by a triangular fuzzy number (TFN) denoted as ${\tilde{W}}_i$. This representation captures the uncertainty and variability in expert judgments, allowing for a more precise evaluation of the importance of each criterion within the decision-making process. By utilizing TFNs, the method effectively incorporates diverse perspectives, enhancing the overall robustness of the risk assessment framework.

The decision matrix was normalized to ensure comparability across different criteria. The normalized value ${\tilde{r}}_{ij}$ for each element was calculated using Equation (2), as follows:

$${\tilde{r}}_{ij}={\displaystyle \frac{{\tilde{x}}_{ij}}{\sqrt{{\sum }_{i=1}^m{\left({\tilde{x}}_{ij}\right)}^2}}},$$

(2)

where ${\tilde{x}}_{ij}$ represents the fuzzy value of the $i$-th alternative concerning the $j$-th criterion.

To determine the fuzzy positive ideal solution (FPIS) and fuzzy negative ideal solution (FNIS), the FPIS ${\tilde{A}}^{+}$ and FNIS ${\tilde{A}}^{-}$ were established according to Equation (3). The FPIS is defined as the best possible performance across all criteria, while the FNIS represents the worst possible performance. These ideal solutions serve as benchmarks for evaluating the performance of each alternative in relation to the criteria set forth in the decision-making process.

$$\begin{array}{cc}\hfill {\tilde{A}}^{+}& =\left(\max {\tilde{x}}_{ij}\right)\hfill \\ \hfill {\tilde{A}}^{-}& =\left(\min {\tilde{x}}_{ij}\right)\hfill \end{array}.$$

(3)

The FPIS signifies the optimal performance for each criterion, whereas the FNIS denotes the least favorable performance. To evaluate the performance of each alternative, the distance from both the FPIS and FNIS was computed using Equation (4). This calculation enables a comprehensive assessment of how closely each alternative aligns with the ideal and non-ideal solutions, facilitating a clearer ranking of the risks involved. By analyzing these distances, decision makers can better understand the relative strengths and weaknesses of each alternative in the context of the established criteria.

$$\begin{gathered} D_i^{+}=\sqrt{{\sum }_{j=1}^{n}d{\left({\tilde{r}}_{ij},{\tilde{A}}^{+}\right)}^2} \\ {D}_i^{-}=\sqrt{{\sum }_{j=1}^{n}d{\left({\tilde{r}}_{ij},{\tilde{A}}^{-}\right)}^2}, \end{gathered}$$

(4)

where $d\left({\tilde{r}}_{ij},{\tilde{A}}^{+}\right)$ and $d\left({\tilde{r}}_{ij},{\tilde{A}}^{-}\right)$ are the distances between the normalized fuzzy values and the FPIS and FNIS, respectively.

The closeness coefficient $C{C}_i$ for each alternative was calculated to assess its relative proximity to the fuzzy positive ideal solution (FPIS), as outlined in Equation (5). This coefficient provides a quantitative measure of how closely each alternative aligns with the ideal solution, enabling decision makers to rank the alternatives based on their performance. A higher closeness coefficient indicates a better fit to the FPIS, facilitating informed choices regarding risk management in GCPs. By analyzing the $C{C}_i$ values, stakeholders can prioritize alternatives more effectively, guiding their decision-making processes.

$$C{C}_i={\displaystyle \frac{D_i^{-}}{D_i^{+}+D_i^{-}}}.$$

(5)

The higher the value of $C{C}_i$, the closer the alternative is to the ideal solution.

The fuzzy weight for each criterion $W_j$ was calculated using the following aggregation method based on the input of multiple experts (Equation (6)):

$$W_j={\displaystyle \frac{1}{n}}{\sum }_{k=1}^n{\tilde{W}}_{jk},$$

(6)

where $n$ is the number of experts, and ${\tilde{W}}_{jk}$ is the fuzzy weight assigned by expert $k$ to criterion $j$.

Finally, the weighted normalized decision matrix was constructed as follows (Equation (7)):

$${\tilde{v}}_{ij}={\tilde{r}}_{ij}\times {\tilde{W}}_j,$$

(7)

where ${\tilde{v}}_{ij}$ represents the weighted normalized value for alternative $i$ under criterion $j$.

2.4. Machine Learning for Risk Ranking

To refine the risk prioritization in GCPs, an EANN was implemented. The EANN is an advanced form of artificial neural network (ANN) that integrates an emotional learning mechanism into the traditional network structure. This allows it to better model complex and dynamic systems by adjusting neuron behavior based on emotional signals. In the context of this study, the EANN was trained using expert-assigned weights and rankings derived from the fuzzy TOPSIS method. This hybrid model allowed for a refined prediction of the risk hierarchy by combining expert judgment with data-driven analysis [65,66,67].

EANN extends the traditional ANN by introducing an artificial emotional component that regulates the behavior of neurons. Each neuron in the EANN can dynamically adjust its activation, learning rate, and output based on inputs and internal emotional signals. This structure enables the network to better handle uncertainties and adapt to changing conditions, which is particularly useful for risk ranking in GCPs where data may be incomplete or uncertain.

Each node in the EANN is responsible for processing data by incorporating hormone-like signals, denoted as H₁, H₂, and H₃, which influence neuron behavior. These hormones are generated dynamically as the network processes input data and adjusts the weight coefficients. The EANN learns through iterative training, during which the hormone coefficients influence the activation function, performance, and output of each neuron [68,69,70,71].

The EANN architecture consists of three primary layers and an input layer, and it receives the weighted criteria and closeness coefficients from the fuzzy TOPSIS output. Hidden layers process the input using neuron nodes influenced by emotional hormones. The output layer produces the final risk ranking as a prediction of risk priority. The output of each neuron in the EANN is governed by Equation (8) [72,73], as follows:

$$Y_i=(\underset{1}{\underbrace{{\gamma }_i+{\sum }_h{\partial }_{i,h}{H}_h)}}\times f\left({\sum }_j[\underset{2}{\underbrace{({\beta }_i+{\sum }_h{\chi }_{i,h}{H}_h)}}\times \underset{3}{\underbrace{({\alpha }_{i,j}+{\sum }_h{\mathsf{\Phi }}_{i,j,k}{H}_h)X_{i,j}}}+\underset{4}{\underbrace{({\alpha }_i+{\sum }_h{\chi }_{i,h}{H}_h)}}]\right),$$

(8)

where $Y_i$ is the output of neuron $i$; $H_h$ represents the emotional hormone affecting the neuron; $\partial ,\beta ,\chi ,\alpha ,\mathsf{\Phi }$ are the weight coefficients; $f$ is the activation function (e.g., sigmoid or ReLU); and $X_{i,j}$ is the input to neuron $i$ from neuron $j$. The hormone values $H_h$ are computed by Equation (9), as follows:

$$\begin{array}{c}{H}_h=\sum _i{H}_{i,h}\left(h=1,2,3\right)\\ H_{i,h}={\mathit{glandity} }_{i,h}\times Y_i\end{array}.$$

(9)

In these equations, the hormones $H_h$ are influenced by the glandity values, which regulate the emotional intensity of each neuron’s response. The hormone factors $H_1,H_2,H_3$ guide the training process by adjusting neuron weights based on the input–output relationship [74,75].

The training of the EANN was conducted using the Levenberg–Marquardt (LM) optimization algorithm. This method, which combines the advantages of both gradient descent and Gauss–Newton methods, is particularly suited for solving non-linear least squares problems. It has been widely used for training neural networks due to its efficiency and speed, especially when dealing with complex datasets like those used in this study. The LM algorithm works by iteratively updating the weight coefficients in the neural network to minimize the error between the predicted outputs and the target values (expert-assigned risk rankings). The key idea behind the LM algorithm is to interpolate between the Gauss–Newton method (for fast convergence near the minimum) and gradient descent (for stability far from the minimum). At each iteration, the weight update is computed by Equation (10), as follows:

$$\mathsf{\Delta }w=-{\left(J^{T}J+\lambda I\right)}^{-1}{J}^{T}e,$$

(10)

where $\mathsf{\Delta }w$ is the update to the weight vector, $J$ is the Jacobian matrix of the network’s error function, $e$ is the error vector (difference between predicted and target outputs), $\lambda$ is the damping factor, adjusted dynamically during training, and $I$ is the identity matrix.

The EANN was initialized with random weights and biases. Hormonal coefficients $H_1,H_2,H_3$ were also initialized based on random values, reflecting the initial emotional states of the neurons. The input data, consisting of the weighted criteria and Fuzzy TOPSIS outputs, were fed into the network. The EANN processed the data through its layers, producing an output for each risk.

In the training process, the EANN was initialized with random weights and biases. Hormonal coefficients $H_1,H_2,H_3$ were also initialized based on random values, reflecting the initial emotional states of the neurons. The input data, consisting of the weighted criteria and fuzzy TOPSIS outputs, were fed into the network. The EANN processed the data through its layers, producing an output for each risk. The difference between the predicted output and the target (expert-assigned rankings) was computed as the error vector $e$. The LM algorithm was applied to update the network’s weights by calculating $\mathsf{\Delta }w$ based on the current error. The damping factor $\lambda$ was adjusted dynamically. When the error decreased, $\lambda$ was reduced, making the update more like the Gauss–Newton method. When the error increased, $\lambda$ was increased, making the update more like gradient descent for stability. The training process continued iteratively, adjusting the weights and hormone coefficients until the error converged to a minimum, or a predefined stopping criterion (such as a small enough error or maximum iterations) was reached.

The input to the EANN consists of the weighted criteria and closeness coefficients from the fuzzy TOPSIS output. These serve as the features $X_{i,j}$ for each neuron.

• Hormonal Influence: During the training process, hormone values $H_h$ adjust the weight coefficients $\partial ,\beta ,\chi ,\alpha ,\mathsf{\Phi }$, influencing the neuron’s behavior. The final output $Y_i$ is calculated for each risk, producing a ranking based on the learned weights and hormone influence.
• Risk Ranking: The predicted risk ranking is derived from the EANN output, providing a hierarchy of risks where a higher $Y_i$ value indicates a more critical risk.

Four scenarios for training the model were adopted in this study. In Scenario A, the overall risk evaluation, the EANN model was trained using a broad dataset encompassing various risks identified through expert assessments. The focus of Scenario B, evaluating material quality risks, was specifically on risks associated with material quality and equipment. In Scenario C, evaluating stakeholder engagement risks, the model was trained on data regarding stakeholder engagement and resistance. Scenario D, which evaluated economic and regulatory risks, concentrated on economic fluctuations and regulatory compliance risks.

The proposed flowchart for the algorithm is presented in Figure 1. The hybrid model enhances decision making in several key ways. First, fuzzy TOPSIS alone ranks risks by processing expert assessments through fuzzy logic, which is useful for capturing ambiguity but does not adaptively refine risk priorities. By integrating EANN, the model not only ranks risks, but also continuously adjusts to patterns in expert-assigned weights, capturing dynamic dependencies between risks. This combination allows for a more robust prioritization that reflects real-world complexities in GCPs.

In response to the inherent complexities and dynamic nature of GCPs, this study integrates an EANN with fuzzy TOPSIS to enhance risk prioritization. While traditional TOPSIS is effective in establishing a fixed ranking of risks based on initial conditions, it lacks adaptability to shifting project environments. EANN adds an adaptive layer to the model by incorporating emotional feedback mechanisms that mimic human-like adjustments in response to changes. This adaptability allows the EANN to dynamically re-evaluate and update risk priorities in real-time, based on evolving project inputs such as material quality shifts, regulatory changes, or stakeholder responses. In this way, the hybrid model provides a robust solution for projects with fluctuating risk profiles, where traditional methods may struggle to account for interdependencies and real-time adjustments. Thus, the inclusion of EANN complements fuzzy TOPSIS by enhancing the model’s responsiveness to complex, interdependent risks, making it highly suitable for the green construction sector.

Fuzzy TOPSIS enables a systematic comparison across multiple criteria by evaluating the closeness of each alternative (risk factor) to an ideal solution. This is particularly beneficial for complex projects like GCPs, where risks vary widely in impact and likelihood. However, standalone fuzzy TOPSIS only offers static ranking without the ability to adapt to changes in risk patterns over time. The EANN component introduces adaptability into the decision-making process. Unlike conventional neural networks, EANN incorporates “emotional” signals or hormone-like factors, which adjust neuron responses based on input complexity. This allows the model to dynamically re-rank risks as more data are introduced or as project conditions evolve, capturing real-time shifts in priorities. The EANN continuously learns from the fuzzy TOPSIS rankings and expert input, adjusting weights to better reflect interdependencies among risk factors. This is especially important in GCPs where risks may be interrelated, making a static model insufficient. The hybrid model combines the ranking stability of fuzzy TOPSIS with the adaptability of EANN, resulting in a system that can respond to both current conditions and potential shifts in risk factors. This hybrid approach supports a more nuanced prioritization of risks, making it more flexible and precise than standalone methods.

The methodology begins with collecting expert evaluations of potential risks in GCPs. We utilized a Likert-scale survey, which gathered responses from 18 experts with four questionnaires. Each risk was rated based on perceived severity and likelihood, providing the input data necessary for risk ranking.

The first stage of risk prioritization involves applying fuzzy TOPSIS in order to handle uncertainties in expert judgments. Fuzzy TOPSIS converts the qualitative risk ratings from the Likert scale into fuzzy values, enabling a more accurate assessment of ambiguous data. Each risk factor is then ranked based on its closeness to the ideal solution, resulting in an initial static prioritization of risks.

After the initial fuzzy TOPSIS ranking, the EANN is introduced to dynamically adapt and refine the prioritization based on changes in input patterns. The EANN operates by simulating an “emotional” feedback mechanism that updates its weighting in response to evolving project data. This enables real-time updates in risk ranking as the model adjusts to patterns and relationships that emerge among risk factors, particularly valuable in the context of green construction, where risks can be highly interdependent and subject to change. The combined output of fuzzy TOPSIS and EANN provides a final, adaptive risk prioritization.

3. Results

Given that the objective of this study was to implement a systematic approach for identifying and ranking the risks associated with GCPs, the statistical population was purposefully limited. It comprised individuals directly involved in or knowledgeable about the specific project under study, including 18 experts in the field.

Based on Table 2, the existing risks were presented to GCP experts and rated using a Likert scale.

Table 4. The importance of the risks of GCPs.

Risk Group	Risk Item	Likert Scale Mean	Conventional and Green Buildings (Number of Expert Opinions)	Green Buildings (Number of Expert Opinions)
RG1	RG11	4.32	7	11
	RG12	4.08	15	3
RG2	RG21	4.41	0	18
	RG22	3.82	0	18
	RG23	3.52	18	0
	RG24	3.01	1	17
	RG25	4.71	9	9
RG3	RG31	4.67	15	3
	RG32	4.71	0	18
	RG33	4.44	0	18
	RG34	4.32	3	15
	RG35	3.59	0	18
	RG36	4.12	2	16
RG4	RG41	4.01	0	18
	RG42	3.89	0	18
	RG43	3.91	1	17
	RG44	3.70	0	18
	RG45	4.15	11	7
RG5	RG51	3.70	16	2
RG6	RG61	4.44	0	18
RG7	RG71	3.94	14	4
	RG72	4.18	9	9
RG8	RG81	3.51	18	0
	RG82	3.81	18	0
	RG83	4.08	18	0

quantitatively shows experts’ opinions on how to assess the importance of risks in GCPs in various groups. The risks associated with conventional and green buildings, as well as the common risks between them, were determined in this questionnaire.

The ranking of the criteria is based on the fuzzy TOPSIS method shown in

Table 5. The ranking of the risk criteria of GCPs.

ID	Criteria	Weight for Ranking
C1	Reaction to risk	0.056
C2	Risk identification	0.09
C3	Uncertainty	0.097
C4	Risk management	0.099
C5	Threat	0.103
C6	Prediction of risk	0.121
C7	Vulnerability	0.123
C8	Outcome	0.137
C9	Risk probability	0.174

. After ranking the criteria, the relative importance of each criterion was calculated.

The normalized weight matrix calculated for different risk groups calculated using the proposed method is shown in

Table 6. Scaleless weighted matrix of risks of different construction industry groups.

Risk Item	C1	C2	C3	C4	C5	C6	C7	C8	C9
RG1
RG11	0.0125	0.0091	0.0055	0.0034	0.0026	0.0019	0.0019	0.0012	0.0008
RG12	0.0097	0.0059	0.0047	0.0035	0.0025	0.0017	0.0019	0.0012	0.0007
RG2
RG21	0.0076	0.0077	0.0063	0.0032	0.0028	0.002	0.0016	0.0012	0.0009
RG22	0.0082	0.0074	0.0067	0.0033	0.0027	0.0021	0.0014	0.0014	0.001
RG23	0.0082	0.0071	0.0065	0.0033	0.0028	0.002	0.0016	0.0013	0.0008
RG24	0.0111	0.0075	0.0064	0.0037	0.0028	0.0022	0.0015	0.0014	0.0008
RG25	0.0079	0.0064	0.0051	0.0035	0.0026	0.0022	0.0016	0.0013	0.0008
RG3
RG31	0.0111	0.0059	0.0061	0.0035	0.0026	0.0021	0.0013	0.0012	0.0009
RG32	0.0088	0.0078	0.0049	0.0034	0.0025	0.0021	0.0017	0.0014	0.0009
RG33	0.0111	0.0042	0.0053	0.0028	0.0022	0.0018	0.0017	0.0013	0.001
RG34	0.0092	0.0076	0.0049	0.003	0.0041	0.00025	0.0027	0.0016	0.0011
RG35	0.0088	0.0078	0.0053	0.0028	0.0023	0.0018	0.0014	0.0011	0.0007
RG36	0.0108	0.0089	0.0065	0.0033	0.0035	0.0023	0.0017	0.0011	0.0009
RG4
RG41	0.0088	0.007	0.0057	0.0033	0.0028	0.0021	0.0018	0.0013	0.0007
RG42	0.0095	0.0092	0.0061	0.0034	0.0017	0.0019	0.0015	0.0015	0.001
RG43	0.0085	0.0095	0.0053	0.0037	0.0021	0.0024	0.0016	0.0012	0.0009
RG44	0.0098	0.0081	0.0057	0.0035	0.0022	0.0021	0.0017	0.0011	0.0009
RG45	0.0088	0.0076	0.0055	0.0031	0.0025	0.0021	0.0014	0.0014	0.0011
RG5
RG51	0.0085	0.0076	0.0059	0.0026	0.0026	0.0021	0.0011	0.0014	0.0008
RG6
RG61	0.0072	0.0087	0.0061	0.0039	0.0023	0.0018	0.0015	0.001	0.0008
RG7
RG71	0.0061	0.0078	0.0049	0.0042	0.0031	0.0014	0.0016	0.0017	0.0008
RG72	0.0072	0.0051	0.0054	0.0031	0.0035	0.0024	0.0014	0.0018	0.0007
RG8
RG81	0.0075	0.0064	0.0057	0.0039	0.0016	0.0014	0.0017	0.0012	0.0008
RG82	0.0074	0.0071	0.0055	0.0039	0.0031	0.0019	0.0017	0.0011	0.0009
RG83	0.0066	0.0056	0.0065	0.0032	0.0031	0.0016	0.0013	0.0016	0.0008

. Figure 2 also shows the final weight calculated for each risk criterion.

Figure 3 shows the evaluation results of the EANN model across four distinct training scenarios, each derived from expert feedback on risks in GCPs. Figure 3a indicated that the model effectively learned the significance of each risk, achieving a decreasing trend in the loss function across the iterations. The initial high loss values gradually declined, demonstrating the model’s ability to refine its predictions based on the collective insights of the expert panel. The loss function depicted in Figure 3b exhibited a notable reduction over the iterations, reflecting the model’s capacity to adapt to the complexities of material-related risks. This scenario highlighted the critical importance of sourcing sustainable materials, and the results reinforced the necessity of stringent quality control measures in GCPs. The loss function trends depicted in Figure 3c demonstrated a consistent decline, indicating that the model effectively captured the intricacies of stakeholder dynamics. The insights from this scenario underscored the need for improved communication and engagement strategies to mitigate risks arising from stakeholder opposition to green practices. Figure 3d revealed that the EANN model successfully learned to identify patterns associated with external economic factors and their impact on project execution. The loss function showed a steady decline, indicating that the model could predict potential challenges in this area effectively.

The relative importance and desirability of each risk have also been calculated in order to rank the risks associated with GCPs. Based on this, Figure 3 depicts the level of desirability and relative importance of risks in GCPs. The risks of green construction are ranked by arranging the value of the options in descending order (

Table 7. Risk item ranking of green construction factors based on the proposed hybrid model.

Risk Item	ID	Ranking
Failure to comply with the standard in green construction	RG11	3
Delay in delivering the equipment to the location	RG12	14
Additional costs due to green construction	RG21	11
The increase in time caused by green construction	RG22	15
Inflation	RG23	13
The effect of currency and interest rate fluctuations on the import of green materials	RG24	22
Incorrect forecasting of market demand	RG25	7
Lack of management experience	RG31	10
Lack of knowledge about technology and green materials	RG32	6
Limited access to green suppliers	RG33	5
Lack of experienced construction companies in green construction	RG34	19
Lack of quantitative evaluation criteria and tools for green performance	RG35	17
Resistance from stakeholders to adopt green ideas	RG36	2
Failure to comply with green building factors	RG41	25
Lack of documents and information for novel green building technologies	RG42	4
Low-quality materials and equipment	RG43	1
Production limitations and new green building technology	RG44	18
Technical complexity	RG45	8
Adverse weather and geological conditions	RG51	24
Lack of clear definition of green building materials	RG61	9
Damages caused by human error	RG71	20
Lack of skilled and experienced professionals	RG72	12
Accidents leading to disability or death	RG81	23
Accidents leading to injury	RG82	21
Lack of safety equipment	RG83	16

).

Table 8. Comparison of the proposed model with standalone fuzzy TOPSIS and traditional decision-making methods for risk assessment in GCPs.

Feature	Proposed Model	Fuzzy TOPSIS	Traditional Decision-Making Methods
Adaptability	The EANN component adjusts dynamically based on changes in risk factors and expert feedback, allowing for continuous adaptation to evolving project conditions.	Ranks risks based on initial input; static model that does not adapt to new data without re-analysis [76].	Methods like AHP or standard MCDM lack the ability to dynamically re-prioritize without extensive recalculations [77].
Complexity Handling	EANN manages complex interdependencies by training patterns in expert data, while fuzzy TOPSIS manages uncertainty in criteria weights. Handles multi-faceted risk data with higher precision.	Effective for initial complexity but cannot account for interdependencies or adapt based on new patterns without re-assessment [78].	Basic MCDM methods handle complexity within set criteria but lack advanced pattern recognition or adaptive capabilities [77].
Management of Uncertainty	Fuzzy logic in TOPSIS allows for the handling of vague inputs, while EANN refines rankings by recognizing uncertain patterns and weighting inconsistencies.	Fuzzy logic allows for uncertainty in input data but lacks further adaptive capabilities to handle ongoing uncertainty [79].	Non-fuzzy MCDM methods struggle with vague data and subjective inputs, and are limited in their treatment of complex uncertainty [80].
Scenario-Based Analysis	EANN’s adaptive training allows it to simulate and analyze different risk scenarios, supporting proactive decision making.	Scenario analysis requires recalculating rankings manually for each scenario, making it impractical in fast-changing environments [81].	Standard methods need multiple re-calculations for scenario analysis, lacking dynamic or predictive capabilities [82].
Ease of Use for Stakeholders	Requires familiarity with neural networks and fuzzy logic, which can be complex, but is manageable with stakeholder training.	Fuzzy TOPSIS is a structured MCDM approach that is relatively easy for stakeholders to understand and implement [83].	Traditional methods like AHP are widely known and easy for stakeholders to grasp but lack the depth and precision of hybrid models [84].
Application Suitability	Best for projects with dynamic risk profiles and complex interdependencies, such as green construction. Adaptable to diverse and evolving conditions.	Suitable for projects with stable risk factors where rankings do not need frequent updating or where limited complexity is present [85].	Suitable for straightforward projects with clearly defined criteria and static risk profiles; limited application in dynamic environments [86].

provides a comparative analysis of the proposed hybrid fuzzy TOPSIS-EANN model against standalone fuzzy TOPSIS and other traditional decision-making methods commonly used in risk assessment for GCPs.

4. Discussion

The following discussion elaborates on the implications of the results, focusing on the key risk factors identified. The first major finding, based on the Likert scale mean values presented in Table 3, is that certain risks were consistently rated as highly critical across the expert panel. Among the 25 risk items evaluated, low-quality materials and equipment (RG₄₃) received the highest importance rating, with an average Likert scale value of 4.71, making it the top-ranked risk in GCPs. This finding is consistent with the previous literature that highlights the detrimental effect of substandard materials on project sustainability and performance [87]. The high rank of this risk underscores the need for stringent quality control measures and better sourcing strategies in green building projects.

Similarly, resistance from stakeholders to adopt green ideas (RG₃₆) was identified as the second most critical risk, also highlighted by the high ranking displayed in Table 4 and visualized in Figure 2. This risk reflects socio-political challenges within the construction industry in Saudi Arabia, where traditional mindsets and lack of awareness hinder the acceptance of sustainable practices. The findings suggest that greater efforts are required to engage stakeholders through education and policy incentives to promote the benefits of green technologies.

In contrast, risks such as adverse weather and geological conditions (RG₅₁) and failure to comply with green building factors (RG₄₁) were ranked much lower, as indicated by their positions in Table 4 and the lower weighting of their relative importance in Figure 3. These results may imply that, while environmental factors are important, they are not perceived as significant obstacles in comparison to issues related to stakeholder engagement and material quality. The lower ranking of weather-related risks may also suggest that GCPs in Saudi Arabia have developed sufficient adaptive measures to address these challenges.

The results from the fuzzy TOPSIS analysis provide additional insights into the weight and significance of each criterion used to evaluate the risks. As shown in Table 5, the criteria risk probability (C₉) received the highest weight (0.174), reflecting its dominant role in the overall risk assessment process. The high importance of risk probability indicates that experts place significant emphasis on the likelihood of risk occurrence, which aligns with common risk management practices where proactive identification and mitigation of probable risks are prioritized.

Other key criteria, such as outcome (C₈) and vulnerability (C₇), also received substantial weights, as visualized in Figure 2. These criteria highlight the focus on the potential consequences and susceptibility of green projects to various threats. The high weighting of these criteria suggests that the experts were particularly concerned with the outcomes of poorly managed risks, including cost overruns, delays, and compromised project sustainability.

On the other hand, criteria such as reaction to risk (C₁) and risk identification (C₂) were assigned lower weights, indicating that, while these factors are important, they were not considered as critical as risk probability and outcomes. This may suggest that there is a stronger emphasis on mitigating the most impactful risks rather than focusing solely on identification or initial reactions to them.

The findings of this study are in line with existing research in the field of green construction risk management. Studies such as those by Polat et al. [41] and Hwang et al. [51] have similarly identified material quality and stakeholder resistance as key risks in green projects. However, this study extends the existing literature by incorporating an innovative hybrid model that combines expert judgment with machine learning to refine risk prioritization.

The application of the EANN has proven effective in enhancing the accuracy of risk rankings, as demonstrated by the alignment between the expert rankings and the model’s predictions. As depicted in Figure 3, the final risk rankings derived from the EANN confirmed the critical importance of risks such as low-quality materials and stakeholder resistance, while providing a more nuanced understanding of lower-ranked risks like adverse weather and technical complexity (RG₄₅). Meanwhile, based on Figure 4, the decreasing trend in the loss function across all scenarios highlights the robustness of the EANN model in training from expert feedback. The results confirm that the model is capable of accurately assessing and prioritizing risks in GCPs based on a comprehensive understanding of the various risk dimensions. Furthermore, the performance of the EANN model in each scenario emphasizes the significance of expert insights in developing effective risk management strategies. By focusing on specific risk categories, the model provides targeted recommendations that can help stakeholders to address the most pressing risks in green construction.

Based on Table 7, the hybrid model combines fuzzy TOPSIS, which manages ambiguity in expert judgments, with an EANN that dynamically adapts based on changing input patterns. This synergy enables the hybrid model to offer enhanced adaptability, complexity handling, and real-time responsiveness.

Standalone fuzzy TOPSIS, while effective in handling initial uncertainties, does not incorporate the dynamic and adaptive characteristics needed in environments with evolving risks, like GCPs. For example, when risk priorities shift due to new data, standalone fuzzy TOPSIS requires complete re-analysis to update the rankings, limiting its effectiveness for real-time decision making. In contrast, the hybrid model’s EANN component captures interdependencies and allows for the rapid re-evaluation of risks as new information becomes available, making it particularly suitable for projects with dynamic risk profiles.

Traditional decision-making methods, such as the Analytic Hierarchy Process (AHP) and other multi-criteria decision-making (MCDM) techniques, provide a straightforward structure that is widely accessible to stakeholders. However, these methods often lack the flexibility and precision needed in complex, multi-dimensional risk assessments. Standard MCDM approaches are limited in their capacity to handle uncertain, interdependent risks and typically lack the adaptability to respond to real-time changes without labor-intensive recalculations. Thus, for GCPs, where risks can vary significantly in both nature and impact, the hybrid fuzzy TOPSIS-EANN model provides a more robust and comprehensive risk assessment framework.

The findings of this study have several practical implications for stakeholders in the construction industry. First, the high ranking of material-related risks suggests that more stringent procurement and quality control processes must be implemented to ensure that GCPs meet the required sustainability standards. Second, the socio-political risks associated with stakeholder resistance highlight the need for more effective communication strategies and engagement mechanisms to secure buy-in from all involved parties. Additionally, the lower ranking of environmental risks, such as adverse weather conditions, may indicate that the existing risk management practices have been successful in mitigating these factors, allowing the focus to shift toward more pressing issues like cost management and regulatory compliance.

5. Conclusions

An integrated fuzzy TOPSIS-EANN model was employed to address the inherent uncertainties in expert judgments and to refine the risk rankings by combining subjective and data-driven methods. The findings of this research provide significant insights into the key risks that GCPs face, and the results offer practical guidance for stakeholders involved in risk management and decision making.

The analysis of expert opinions highlights that low-quality materials and equipment were identified as the most critical risk, with a Likert scale mean of 4.71. In this regard, quality control and the selection of certified, sustainable materials are essential to mitigate this risk and ensure that the project adheres to green building standards. Similarly, stakeholder resistance to adopting green ideas, which received the second-highest ranking, underscores the need for stronger engagement strategies to overcome resistance from individuals or groups who may be unfamiliar with or opposed to sustainable construction practices. The fuzzy TOPSIS analysis further demonstrated that risk probability received the highest weight (0.174), indicating that experts prioritized the likelihood of risk occurrence above other criteria. This suggests that the risk management strategies in GCPs should focus on identifying and mitigating the most probable risks early in the project lifecycle. Moreover, criteria such as outcome and vulnerability were also weighted significantly, with values of 0.137 and 0.123, respectively, highlighting the importance of addressing the potential consequences and weaknesses associated with green construction risks. In terms of practical implications, the results suggest that project managers and decision makers should implement more comprehensive risk management frameworks that focus on high-probability and high-impact risks, such as material quality and stakeholder resistance.

Despite the robustness of the hybrid fuzzy TOPSIS and EANN model, some limitations should be noted. The relatively small sample size of experts may have constrained the diversity of perspectives, although their extensive experience in the field did provide valuable insights. However, more experts were invited to complete the survey, but a limited number cooperated. Meanwhile, the community of experts in the field of GCPs in Saudi Arabia is limited due to its nascent nature. In addition, the risks associated with green construction are dynamic and may evolve as new technologies and policies emerge. This study focused on current risks, but continuous updates to the risk model would be necessary to keep it relevant over time. Future studies could expand the expert panel to include a broader range of stakeholders, such as government officials, contractors, and investors, to provide a more comprehensive view of the risks associated with GCPs. Future research could apply the proposed model to other developing countries to assess whether similar risks and priorities emerge in different socio-economic, political, and environmental contexts.