Applications of machine learning and artificial intelligence in construction project cost prediction: a scientometric analysis and qualitative review

Abstract

Accurate construction project cost prediction directly affects investment decision-making, resource allocation efficiency, and project risk management. Under increasingly complex project environments, traditional estimation methods face limitations in handling high-dimensional data, nonlinear relationships, and dynamic fluctuations. Accordingly, this study conducts a scientometric analysis and qualitative review of the application of machine learning and artificial intelligence in construction cost prediction. The PRISMA framework was adopted for systematic literature retrieval and screening to ensure the transparency and reproducibility of the dataset construction process. Web of Science and Scopus were used as the primary data sources, from which 138 relevant articles were ultimately selected. Subsequently, scientometric analysis, knowledge graph visualization, and qualitative analysis were integrated to systematically examine the research evolution, knowledge structure, model application characteristics, and future development trends in this field. Findings indicate that since 2021, the field has entered a rapid expansion phase, with research focus shifting from early single-model validation toward integrated, deep, and hybrid approaches. ML and AI methods more effectively capture complex relationships influencing construction costs and demonstrate superior predictive accuracy and scenario adaptability compared with traditional methods. However, practical application remains constrained by unstable data quality, limited cross-regional generalization, and low model interpretability. By integrating scientometric results with content analysis, this study not only elucidates the evolution of research hotspots but also distills the mechanisms by which project attributes, design decisions, resource price fluctuations, and external environmental disturbances collectively drive construction costs. The findings provide a comprehensive review framework and offer guidance for optimizing predictive models, advancing multi-source data integration, and supporting intelligent decision-making in construction management.

1 Introduction

Cost forecasting in construction projects is one of the most critical and challenging aspects of project management. Accurate early-stage cost estimation directly influences owners’ investment decisions, contractors’ bidding strategies, project financing feasibility, and overall project success. Throughout the full lifecycle of a construction project, costs encompass multiple dimensions, and both underestimation and overestimation can lead to serious consequences. Underestimating construction costs may trigger cash flow disruptions, schedule delays, or even project failure, whereas overestimation can result in abandoning high-quality projects due to budget constraints or misallocation of resources. This issue is particularly pronounced in the current global construction environment, characterized by volatile material prices, unstable supply chains, and persistent geopolitical and inflationary pressures. Consequently, obtaining precise and reliable cost forecasts has become a key determinant of project competitiveness and sustainability (Jafari and Mousavi, 2026).

Traditional cost estimation methods mainly include expert judgment, unit cost, parametric models, and quantity survey-based approaches (Niazi et al., 2006). These methods provide reference value when project designs are relatively mature, drawings are complete, and historical data are stable. Expert judgment relies on the experience and intuition of senior cost engineers, making it suitable for rapid conceptual-stage estimates. The unit cost method calculates total costs by multiplying quantities by historical unit prices, offering a systematic approach. Parametric models leverage macro-level parameters such as building area, volume, or functional units to generate quick estimates. However, these conventional methods exhibit notable limitations. First, they heavily depend on the estimator’s subjective experience, introducing human bias and limiting reproducibility. Second, they adapt poorly to incomplete early-stage project information; when design depth is insufficient or scope boundaries are unclear, estimation errors tend to be high. Third, they struggle to capture dynamic uncertainties such as material price fluctuations, labor market changes, or site-specific geotechnical conditions, leading to persistent deviations between predicted and actual costs. Fourth, these methods are time-consuming and inefficient, especially in medium- to large-scale complex projects, where comprehensive quantity calculation and verification can take several weeks or even months (Cheng and Hoang, 2014).

In recent years, with substantial improvements in computing power, accumulation of extensive historical project data, and breakthroughs in algorithmic theory, machine learning (ML) and artificial intelligence (AI) have demonstrated exponential growth in construction cost forecasting applications (Chen and Guestrin, 2016). The concept of machine learning can be traced back to Arthur Samuel’s checkers program in the late 1950s, while modern frameworks of supervised learning, unsupervised learning, and reinforcement learning gradually matured in the 1990s, particularly with the development and optimization of classic algorithms such as support vector machines, decision trees, and neural networks. Entering the 21st century, the rise of deep learning and rapid advancements in gradient boosting tree algorithms have significantly enhanced models’ abilities to process high-dimensional, nonlinear, and noisy data. An et al. first applied support vector machines (SVM) for conceptual cost estimation in construction projects, demonstrating the feasibility of ML in early-stage cost prediction (An et al., 2007). Oztekin and Masterson further proposed an integrated cost estimation framework based on predictive analytics, addressing the limitations of traditional methods under multivariable and qualitative conditions (Oztekin and Masterson, 2018). In construction engineering, AI/ML applications have evolved from early artificial neural networks (ANN) in the 2000s to support vector regression (SVR) (Gunaydın and Dogan, 2004), random forests (RF), gradient boosting decision trees (GBDT) (Alshboul et al., 2022a), deep neural networks (DNN), long short-term memory networks (LSTM), and various ensemble or hybrid models.

Comparative studies between ANN and case-based reasoning (CBR) methods further promoted the development of intelligent cost prediction approaches (Kim et al., 2005). Compared with traditional methods, data-driven ML/AI models offer significant advantages. They can automatically learn complex nonlinear relationships without predefined functional forms, capturing interaction effects and implicit patterns that traditional parametric models cannot describe (Shamim et al., 2025). They are more robust to incomplete early-stage data; through feature engineering and historical project matching, they can provide reasonable range predictions even with limited input information. Prediction speed is extremely fast, as trained models can generate forecasts within milliseconds, facilitating rapid scenario comparisons during bidding. ML/AI models can integrate multiple data sources, combining structured tabular data such as floor area and number of stories with time series, thereby enhancing information utilization (Tang and Shide, 2025). Moreover, interpretability has gradually improved, with post hoc analysis methods such as SHAP, LIME, and PDP making “black box” model decision paths more transparent for practitioners (Molnar et al., 2020).

Despite some existing reviews on ML applications in construction cost forecasting, gaps remain. Many studies focus on single algorithms or specific project types, offering limited systematic comparison and synthesis across different ML methods, making it difficult to reflect the performance and applicability of models under diverse scenarios (Marzouk et al., 2024; Elmousalami, 2020). Furthermore, some reviews emphasize algorithmic or model-level summaries while lacking systematic analysis of data sources, feature selection, and specific application contexts, limiting the practical guidance of conclusions (Zhou et al., 2024). Existing reviews also tend to concentrate on traditional ML techniques, with less attention to rapidly developing deep learning models, Building Information Modeling (BIM) data integration, and multi-source data-driven cost prediction methods, thus failing to fully capture recent advancements in the field (Fu et al., 2024). Additionally, literature search scope and database coverage in some reviews are limited, reducing representativeness and comprehensiveness (Wang et al., 2024). Therefore, a broad and systematic review of AI and ML applications in construction cost forecasting is warranted to identify current research hotspots, technological trends, and potential future directions.

Accordingly, this study aims to provide a comprehensive review of AI and ML applications in construction cost prediction. Relevant literature was collected through database searches, followed by scientometric analysis to quantify development patterns and main research themes in this domain. Using VOSviewer, keyword co-occurrence, research hotspots, and academic collaboration networks were visualized to reveal the knowledge structure and evolution of the field (Al Husaeni, 2023). A qualitative review was then conducted to systematically analyze and compare the types of ML models applied, data sources, feature variables, and specific application scenarios, summarizing the characteristics and development trends of various models in cost prediction, and highlighting existing limitations and challenges to provide theoretical references and research directions for future studies (Jin et al., 2019).

The remainder of this paper is organized as follows: Chapter 2 details the research methodology, including literature search strategies, scientometric analysis methods, and qualitative analysis workflow. Chapter 3 presents the scientometric analysis results, focusing on development trends and visualizing the knowledge network and academic collaborations among research themes. Chapter 4 builds on these results with qualitative analysis, systematically summarizing data sources, key influencing factors, problem classification, and interrelationships. Chapter 5 evaluates the potential of AI and large models in analyzing complex relationships, discussing current limitations and future directions. Chapter 6 concludes the study, summarizing key findings and providing corresponding research insights.

2 Research methods

Scientometric analysis, as a powerful quantitative research approach, can reveal deep literature connections that are difficult to capture through traditional qualitative reviews (Wang J. et al., 2022). Compared with conventional citation-based metrics, scientometrics integrates multiple tools and datasets to identify underlying patterns in scholarly output and uncover potential insights and emerging trends within a discipline (Hosseini et al., 2018). In this study, scientometric analysis was adopted as the primary research method, while the PRISMA framework was employed for systematic literature retrieval and screening to ensure the transparency and reproducibility of the dataset selection process. In addition, qualitative interpretive analysis was used as a complementary analytical component to support discussions on methodological evolution, model characteristics, and research limitations. This study adopts a widely recognized “three-stage approach” for structured analysis (Elmousalami, 2020; Zhang et al., 2024), as illustrated in Figure 1.

In Stage 1, relevant literature was retrieved from the Web of Science (WOS) and Scopus databases using the keywords shown in Figure 1. The PRISMA framework was applied for multiple rounds of screening and normalization (Moher et al., 2010), ultimately selecting high-quality core papers that meet the study criteria as the analysis foundation. Stage 2 focuses on interpreting development trends in the field using MS Excel and constructing co-authorship, keyword, and citation networks with VOSviewer, transforming literature associations into intuitive knowledge maps. Stage 3 employs multidimensional analysis to reveal technological hotspots and the maturity of machine learning applications in cost forecasting. This stage not only systematically delineates the development trajectory and knowledge structure of the field but also provides scientific guidance for research institutions and decision-makers in optimizing resource allocation and promoting intelligent construction cost management. Collectively, these three stages create a complete logical loop from data acquisition to insight generation.

2.1 Data collection

Following the PRISMA framework, a structured multi-stage screening process was implemented (Jafari and Mousavi, 2026). In the initial retrieval stage, WOS and Scopus were used to search for studies on machine learning and AI applications in construction project cost forecasting. Four groups of keywords were used: (1) construction; (2) cost; (3) forecasting, estimation, or prediction; (4) machine learning, artificial intelligence, or data-driven methods. Literature retrieval was conducted using the following Boolean search expression: (“Construction”) AND (“Cost”) AND (“Forecast” OR “Forecasting” OR “Prediction” OR “Predict” OR “Estimate” OR “Estimation”) AND (“Machine Learning” OR “Artificial Intelligence” OR “ML” OR “AI” OR “Data-driven”). The search scope was limited to the title, abstract, and keyword fields, and only English-language original journal articles published between 1994 and 2025 were retained. A total of 1,944 relevant studies were initially identified. After removing 323 duplicate records, 1,621 articles remained for subsequent screening.

Following this rigorous screening, 138 journal articles were identified as the core corpus for this review. Under the guidance of the PRISMA framework, a systematic literature retrieval and screening process was established. The selected articles were annotated and their full records and citation information exported for VOSviewer analysis, enabling the visualization of networks to help readers quickly identify research hotspots, collaboration patterns, and key connections in construction cost prediction.

2.2 Scientometric analysis

This study combines scientometric analysis with visualization techniques to present complex literature data intuitively. The approach facilitates rapid identification of collaboration networks, emerging trends, and knowledge structures in the application of ML and AI to construction cost forecasting, providing actionable insights for project decision-makers and relevant government agencies regarding resource allocation and policy evaluation.

MS Excel and VOSviewer were used as analytical tools. MS Excel was employed to compile statistics and generate trend charts of publication volumes across years, revealing the development trajectory and stage characteristics of ML/AI research in construction cost forecasting. VOSviewer was used for scientometric network analysis, focusing on co-authorship among researchers, countries/regions, and institutions, as well as keyword co-occurrence networks. In the networks constructed using VOSviewer mapping and clustering, node size indicates weight, line thickness represents the strength of association, and color reflects cluster membership. This visualization clarifies academic interactions among core authors and thematic relationships among keywords (Cobo et al., 2011; Van Eck and Waltman, 2010). The combined methodology not only maps the knowledge domain of construction cost forecasting but also highlights core research hotspots and knowledge dissemination pathways, supporting both quantitative and qualitative assessment of the field and informing future research directions (Zhai et al., 2020).

2.3 Qualitative analysis

The qualitative analysis stage aims to systematically synthesize and interpret the selected literature, focusing on model architectures, data processing strategies, and predictive performance of ML and AI approaches in construction cost forecasting. Key developments, limitations, and recommendations for future research are summarized.

Building on the scientometric and visualization results, the qualitative review systematically examined primary data sources, feature selection, and preprocessing approaches used in cost prediction studies. By integrating earlier scientometric findings with relevant predictive theory, the analysis identified core factors influencing prediction accuracy, including input variable types and models’ intrinsic mechanisms for capturing nonlinear relationships, handling noisy data, and managing uncertainty. Different models were objectively evaluated with respect to applicability across construction project lifecycle stages, computational resource requirements, and generalization performance. Classification based on typical project data characteristics and scenarios allowed assessment of models’ ability to identify sources of cost deviations, provide overspending warnings, and maintain predictive robustness, offering a theoretical foundation for enhancing practical utility and decision support. Additionally, mainstream predictive algorithms were reviewed, emphasizing research advances and technical breakthroughs in improving accuracy, reducing systematic bias, and enhancing model interpretability.

3 Analysis and discussion

3.1 Publication output

The number of publications in a given year is a fundamental indicator for assessing the dynamic evolution and scholarly activity within the field of cost prediction (Larsen and Von Ins, 2010). By analyzing the annual publication counts of the 138 core articles included in this review, the development trajectory and growth trends of ML and AI applications in construction cost forecasting can be clearly illustrated. The starting year of this analysis is 1994, based on carefully screened literature. The annual publication distribution from 1994 to 2025 is shown in Figure 2.

As shown in Figure 2, only a few publications appeared between 1994 and 2008, indicating that ML and AI applications in construction cost forecasting were still in the exploratory and theoretical research phase during this period. Between 2009 and 2020, the number of publications increased gradually, with some data accumulation, signaling a marked rise in research activity; a total of 28 papers were published in this period, accounting for 20.29% of the total. The attention of the construction management field toward intelligent cost estimation steadily increased after 2009.

From 2021 to 2025, the publication volume experienced exponential growth with a sharply increasing slope, reaching 41 papers in 2025, which represents 76.81% of the total publications. This phase clearly reflects an unprecedented peak in the application of ML and AI for construction cost estimation. Notably, the publication count in 2023 shows a significant decline. This reduction may be attributed to the severe global construction supply chain fluctuations during 2021–2022, which rendered historical cost data less reliable, forcing researchers to revise predictive features and model parameters. This reconstruction of empirical data is reflected in the temporary low in publication numbers for 2023, while simultaneously laying the groundwork for the rebound observed from 2024 onward.

3.2 Co-authorship

Studying scientific collaboration within a specific field not only facilitates knowledge acquisition and broadens research perspectives but also reveals the intrinsic mechanisms of knowledge production and dissemination (Newman, 2004). Based on the literature database compiled in this study, co-authorship network analysis can effectively identify core authors, collaborative clusters, and cross-regional collaborations, thereby evaluating the field’s innovation vitality and degree of internationalization (Zhou et al., 2025). In the rapidly developing interdisciplinary area of ML and AI applications in construction cost forecasting, multidisciplinary integration has become a key driver for improving model accuracy and practical implementation. Therefore, the following analysis focuses on co-authorship networks, detailing collaboration patterns among researchers, countries/regions, and institutions, and revealing the dynamics of knowledge flow to support future research and managerial decision-making.

3.2.1 Researcher co-authorship

VOSviewer was employed to construct an author co-authorship network to reveal collaboration patterns among core researchers in ML and AI applications for construction cost forecasting. A minimum document threshold of 2 was applied, resulting in 65 core authors selected from 368 contributors. The resulting visualization is shown in Figure 3. In the network, node size represents the number of publications, line thickness indicates collaboration strength, and color gradients reflect publication years. The overall network is moderately dispersed, with several tight clusters and bridging nodes, illustrating that collaborations are primarily regional, with emerging international connections.

Figure 3 shows multiple small but high-intensity clusters. The network exhibits a moderate level of centralization, with a few prolific authors acting as critical bridges. The temporal color distribution highlights “Cheng, Min-Yuan” as the most prominent node, with the largest size and dark blue color, located on the right side of the network. Strong connections exist between this author and Yu-Wei Wu, forming a blue cluster. This cluster has been active since 2016, spanning the longest period, and its publications have received up to 456 citations, indicating foundational contributions to the field. Cheng’s high-impact work on evolutionary fuzzy neural networks, support vector machines, and hybrid intelligent models for construction cost estimation, including early ANN optimization models, established a key knowledge base. “Gransberg, Douglas D” represents a small blue cluster, reflecting early North American contributions to cost management theory.

Another notable node, “Wang, Jun,” forms a close collaboration group with Ziyi Qu and Nicholas Chileshe and has indirect links with Joon-soo Kim, demonstrating a bridging role in cross-regional collaboration. Since 2017, Wang has published four papers in cost prediction, reflecting the active participation of Asian scholars and emphasizing research on ML algorithm optimization and practical deployment of cost prediction models. These patterns suggest that institutional collaborations can extend toward Asia and the Middle East.

3.2.2 Country co-authorship

A country-level co-authorship network was constructed using VOSviewer to examine international collaboration patterns. To ensure rigor, this study adjusted the original dataset by consistently classifying Taiwan Province as part of China. Applying a minimum document threshold of 2, 22 countries were included in the analysis, resulting in the visualization shown in Figure 4. Node size represents collaboration activity, line thickness reflects collaboration strength, and color gradients indicate the average publication year. The network exhibits a multi-centered, radial structure, with the United States and China as central hubs extending connections to the Middle East, Asia, and other regions.

China emerges as the largest node and a core contributor, maintaining extensive collaborations with multiple countries/regions. The United States is another key hub, playing a critical role in cross-regional collaborations. The network shows clear regional clustering: Southeast and South Asian countries, represented by India and Malaysia, form tightly connected clusters with yellow-colored nodes, indicating particularly active collaboration during 2023–2025, marking them as recent research hotspots. Middle Eastern countries, centered on Iraq and Iran, act as bridges connecting Southeast Asia with major research hubs such as China and the United States. In contrast, nodes representing the United States, the United Kingdom, and Australia appear more blue-purple, reflecting collaborations primarily formed between 2020 and 2022, with relatively lower recent activity compared to Southeast Asia and the Middle East. These patterns suggest that international collaboration is increasingly oriented toward emerging market countries, with both regional cohesion and cross-regional exchanges driving the field’s development. The regional concentration phenomenon indicates that ML- and AI-based cost prediction models are highly dependent on the data environment of specific regions. Differences across regions in project characteristics, market conditions, and data quality may lead to a decline in model generalization performance when applied cross-regionally. Therefore, future research should further strengthen cross-regional validation and multi-source data fusion to improve the applicability and robustness of these models.

3.2.3 Institution co-authorship

Analyzing institutional co-authorship is essential for understanding collaborative patterns and identifying core research forces in ML and AI applications for construction cost forecasting (Yevu et al., 2021). VOSviewer was used to construct an institutional co-authorship network. With a minimum document threshold of 2, 30 core institutions were selected from 225 candidates. Node size represents publication activity, line thickness indicates collaboration strength, and the overall layout is relatively dispersed, showing multiple regional clusters and highlighting predominantly intra-institutional collaboration with limited cross-regional integration.

Figure 5 illustrates collaboration among the most influential institutions. National Taiwan University of Science and Technology emerges as the most active node, located at the bottom of the network with a purple color. Despite being the primary contributor, its connections to other institutions are minimal, indicating high output but limited external collaboration. Another significant cluster, centered on Tarbiat Modares University and Middle East Technical University, with green-colored nodes, is linked to Cihan University Erbil and Lebanese French University. This cluster has high collaboration intensity and has been active since 2020, focusing on cost uncertainty prediction for complex projects such as tunnels and infrastructure, reflecting the Middle East’s regional leadership and application-driven focus (Mostofi et al., 2025; Abed et al., 2022).

The collaboration chain centered on Qingdao University of Technology exhibits strong international integration, connecting closely with Bond University and the University of South Australia, which further links to Birmingham City University in the United Kingdom. Node colors indicate that Qingdao University’s collaborations are concentrated in recent years, representing the most active hub in this chain and highlighting its bridging role in international collaboration. In contrast, nodes such as Cairo University, Cracow University of Technology, and Georgia Institute of Technology are relatively isolated, focusing on theoretical frameworks or specific scenarios, with earlier publication dates and limited impact, suggesting that the European and North American institutional networks remain underdeveloped (Wang and Ashuri, 2017; Chen et al., 2025a). These findings provide quantitative guidance for identifying high-potential partners, promoting cross-regional research collaboration, enabling dataset sharing, standardizing models, and supporting industry adoption.

3.3 Keyword co-occurrence

This study employed VOSviewer to construct a keyword co-occurrence network (Zang et al., 2025), aiming to reveal the core thematic structure and research hotspots in the field. The network was based on author keywords as well as terms extracted from titles and abstracts, with a minimum occurrence threshold set to 4, resulting in the visualization shown in Figure 6.

In ML- and AI-based construction cost forecasting, the keyword co-occurrence network reveals the core thematic structure, evolution pathways, and emerging research areas. Node size represents term frequency, while link strength indicates co-occurrence intensity. The network exhibits a highly interconnected hub structure, with central keywords such as “machine learning,” “neural networks,” and “artificial intelligence,” forming multiple dense clusters. Overall, the field has evolved dynamically from early validation of classical algorithms toward recent multimodal integration and application optimization.

3.4 Citation analysis

Journal co-citation analysis was conducted to reveal the underlying structure of knowledge in ML and AI applications for construction cost forecasting, helping to identify core sources and track the evolution of research themes (Xu et al., 2022). Using VOSviewer, a minimum threshold of two citations per journal was applied, resulting in 21 core journals selected from 55 sources. Figure 7 presents the journal co-citation network, where node size reflects citation frequency or impact, and line thickness indicates co-citation strength. Cluster analysis shows clear interdisciplinary integration among construction management, computational intelligence, and sustainable building research.

Figure 7 highlights five core journals in the field. Early major journals, including Automation in Construction and Journal of Construction Engineering and Management, appear in blue-green nodes. Journals such as International Journal of Construction Management, Engineering, Construction and Architectural Management, and Buildings appear in yellow nodes, indicating more recent activity. Among these, International Journal of Construction Management exhibits the highest recent activity. Another important cluster consists of AI cross-application journals, including Expert Systems with Applications, Advanced Engineering Informatics, and Journal of Computing in Civil Engineering. These journals focus on algorithm and intelligent system applications and were most active in the early stages, reflecting the integration of expert systems, neural networks, and other AI techniques into civil engineering research.

4 Qualitative discussions

Building on the preceding scientometric analysis, this section provides a systematic qualitative review of the literature to examine the characteristics, methodological development, and application trends of ML and AI in construction cost forecasting. Previous studies have highlighted that combining scientometric analysis with qualitative synthesis allows researchers to identify macro-level trends while gaining deeper methodological and technical insights, ultimately forming a more systematic knowledge framework (Zang et al., 2025).

By integrating information on model types, data sources, research methods, evaluation metrics, and study limitations, it is evident that current ML and AI research in construction cost prediction primarily focuses on the following areas: research methods and model types, data sources and characteristics, cost-influencing factors, model performance evaluation approaches, and research limitations and future directions.

4.1 Research methods and model types

Traditional statistical methods have limitations in handling the nonlinearities inherent in construction cost data, whereas data-driven models such as machine learning and deep learning have emerged as central research trends. To clarify the structure and technological trends of current methodologies, the ML and AI models in the reviewed literature were categorized and analyzed, as summarized in Table 1.

With improvements in computational power, machine learning (ML) techniques have been increasingly used for construction cost forecasting. Common approaches include artificial neural networks (ANN), support vector machines (SVM), decision trees, and ensemble-based methods. ANNs, inspired by biological neural structures, are capable of learning nonlinear relationships between variables and are therefore suitable for multivariable prediction problems (Padala and Goyal, 2025). Despite challenges such as difficult parameter tuning and limited interpretability, many studies have applied ANN models to historical project data to link project features with cost outcomes, achieving relatively good prediction results (Yun, 2022). SVMs are another widely used method, which work by finding optimal hyperplanes for regression or classification in high-dimensional feature spaces. They generally perform well on small and medium datasets with strong generalization ability, although their efficiency decreases when dealing with very large datasets.

In more recent studies, ensemble learning methods such as Random Forest, Gradient Boosting, AdaBoost, and XGBoost have gained increasing attention. These models combine multiple weak learners to form a more robust predictor, improving accuracy while reducing overfitting, and are therefore more stable in practice. At the same time, deep learning methods such as DNN, CNN, and LSTM have been introduced to handle more complex or sequential data. They are able to automatically learn feature representations from raw data, reducing the need for manual feature engineering and often improving prediction performance. However, their application is still limited by the need for large datasets and relatively high data requirements in construction cost prediction tasks.

Although machine learning and artificial intelligence models generally outperform traditional statistical methods, significant differences still exist among models in terms of predictive accuracy, interpretability, data requirements, and engineering applicability. Models such as ANN, DNN, and LSTM often achieve higher prediction accuracy, but they rely heavily on large volumes of high-quality data and suffer from the “black-box” problem, resulting in weak interpretability. SVM shows strong generalization ability under small-sample conditions, but its training efficiency becomes limited when applied to large-scale datasets. Ensemble learning methods such as Random Forest and XGBoost achieve a relatively good balance between accuracy and stability; however, they still depend on effective feature engineering and may be affected by regional data bias.

Therefore, there is currently no single optimal model suitable for all engineering scenarios, as clear trade-offs remain among accuracy, transparency, and applicability. Based on this, this study further develops a comparative framework, as shown in Table 2, of different machine learning and artificial intelligence models to improve the practical reference value for model selection.

4.2 Data sources and data characteristics

4.2.1 Data sources

In machine learning model development, data quality and scale play critical roles in determining model performance. Existing studies indicate that most construction cost prediction research primarily relies on historical project data as training samples. These datasets typically include basic project information and key cost-related variables, such as building area, project type, construction duration, material costs, labor costs, and project complexity.

4.2.2 Feature processing

To improve the accuracy of construction cost prediction models, most studies conduct multi-dimensional data preprocessing prior to modeling. Core preprocessing steps include data cleaning, handling missing values, detecting outliers, and data standardization. Beyond these foundational steps, feature selection serves as a critical extension, often employing methods such as correlation analysis, principal component analysis (PCA), genetic algorithms, and random forests to identify key variables, optimize data quality, and enhance model performance.

With advances in digital technologies, emerging data sources such as Building Information Modeling (BIM), the Internet of Things (IoT), and big data platforms have been introduced, providing rich and dynamic information to support predictive models. This shift enables research to move from traditional static analysis toward multi-source data integration and intelligent forecasting. Table 4 summarizes common data preprocessing methods and their intended purposes.

Handling missing values is a fundamental step, commonly using mean or median imputation. Outlier detection often employs Z-score or interquartile range methods, with subsequent deletion or correction to reduce extreme value interference; however, differences in outlier thresholds can affect outcomes. Data scaling is typically achieved through standardization or normalization (e.g., Min-Max scaling, Z-score), improving training efficiency and model adaptability. Transformation methods are also applied to adjust skewed distributions.

Feature engineering is central to enhancing model performance, including correlation analysis, PCA, and other dimensionality reduction techniques to select key variables, mitigate multicollinearity, and reduce noise. One-hot encoding and VIF checks are common approaches. Data augmentation techniques address small sample sizes, while methods like SMOTE handle class imbalance in classification tasks.

Overall, current preprocessing practices reflect progress in data quality improvement and feature optimization, promoting standardized and refined construction cost prediction modeling. Nevertheless, the optimal combination of preprocessing methods and their precise impact on model performance remain areas for further investigation, providing directions for future research.

4.3 Key influencing factors in cost prediction

This study systematically reviews high-frequency factors reported in the literature and synthesizes their roles, transmission mechanisms, and application contexts in the construction cost formation process. The analysis indicates that cost-driving factors in existing research are inherently multi-dimensional and coupled, encompassing project attributes, technical specifications, resource price fluctuations, external economic conditions, and site-specific implementation constraints. Table 5 summarizes the key influencing factors.

Qualitative analysis indicates that influencing factors in construction cost prediction are complex, interrelated, and dynamic. Project-scale characteristics, spatial form, and technical complexity define the baseline cost, forming the structural foundation for predictive modeling. Early-stage design decisions act as front-end control variables, establishing basic cost boundaries and producing a “decision upfront, realization later” dynamic.

Resource prices and macroeconomic fluctuations continuously shift the cost baseline, while site conditions, construction environment, and market competition further enhance variability, leading similar projects to exhibit significant cost differences under different implementation scenarios. Consequently, construction cost is not a static outcome determined by a single variable, but a dynamic product of interactions among intrinsic project attributes, resource constraints, and external environmental factors.

This observation underscores the advantage of ML and AI in cost prediction. Beyond stronger fitting ability, these methods overcome the limitations of traditional approaches that rely on linear relationships or single mechanisms. They are capable of capturing complex, nonlinear interactions, cross-variable dependencies, and temporal propagation patterns among multiple factors, aligning more closely with the true mechanisms underlying construction cost formation.

4.4 Model performance evaluation

The evaluation of construction cost prediction models requires a multi-indicator, multi-level approach tailored to the characteristics of the prediction task and the data. Existing studies generally assess model performance using two core strategies: error quantification and generalization capability verification.

For error measurement, commonly used metrics include Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). These metrics are widely applied due to their simplicity and clear physical interpretation. RMSE, in particular, is sensitive to extreme errors, making it well-suited for complex projects with large cost fluctuations. For instance, Ali Z.H. et al. employed RMSE to validate the performance of an XGBoost model, demonstrating its effectiveness in identifying extreme cost deviations and achieving a high-precision prediction with (Ali and Burhan, 2023).

Mean Absolute Percentage Error (MAPE) standardizes predictions to remove dimensional effects, providing an intuitive measure of relative deviation between predicted and actual costs. This metric is particularly useful for comparing models across projects of varying scale. Salahaldain Z. et al. used MAPE to standardize the evaluation of prediction accuracy between SVM and ANN models (Salahaldain et al., 2023). To address the non-normal distribution often observed in cost data, some studies incorporate Mean Absolute Deviation (MAD) or Median Absolute Deviation (Med AD) to reduce the influence of outliers and enhance robustness (Moghayedi and Windapo, 2023).

Generalization capability is typically assessed according to data characteristics and study design. For static cross-sectional data, K-fold cross-validation (K-fold CV) randomly partitions datasets into training and validation subsets and iteratively evaluates model performance, mitigating overfitting from a single data split. Ten-fold CV is widely adopted as it balances bias and variance (Berawi et al., 2025). Shen Y. applied this method to compare the generalization performance of SVM and KNN regression models for construction cost estimation (Shen, 2025).

For time-series cost data, time-series split validation preserves chronological order, simulating the “future data unknown” scenario in real project forecasting. This approach aligns with the dynamic evolution of construction costs over a project lifecycle (Wang and Ashuri, 2017). Some high-sample studies additionally use leave-one-out cross-validation (LOOCV) to maximize data utilization, though computational costs limit its application to small-sample, specialized cost prediction tasks (Sanni-Anibire et al., 2022).

Overall, current performance evaluation practices combine foundational error metrics with scenario-based generalization assessments, providing a unified framework for comparing the effectiveness of different ML and AI models in construction cost prediction.

5 Current applications and future outlook

5.1 Application progress

ML and AI applications in construction cost prediction have progressed from theoretical validation to practical implementation, forming a technical ecosystem that spans the full project lifecycle and adapts to diverse scenarios.

In early-stage project decision-making, rapid estimation tools based on ANN, SVM, and related models are widely used in conceptual design and bidding stages. By integrating core parameters such as building area and structural type, these tools can deliver minute-level cost estimates with prediction errors typically controlled within 10%–15%, significantly improving decision efficiency.

For specialized projects such as residential buildings, highways, and tunnels, domain-adapted models have emerged. Examples include GRU/LSTM-based highway cost index forecasting models and high-rise building cost estimation systems integrated with BIM attributes. These models have been validated in infrastructure projects across Egypt, South Korea, and China, demonstrating strong scenario adaptability (Alzara et al., 2025; Park and Yun, 2023).

In dynamic cost control and risk warning, ensemble and hybrid models overcome the limitations of traditional static predictions. Hybrid models combining XGBoost with simulated annealing and EVM-based completion cost prediction frameworks integrate real-time data such as construction progress and resource price fluctuations, enabling early risk detection and reducing prediction errors by more than 50% compared with conventional methods (ForouzeshNejad et al., 2024; Narbaev et al., 2024).

Moreover, interpretable AI techniques enhance model transparency. Tools such as SHAP and LIME have been applied in green building cost prediction, not only improving accuracy but also enabling decision-makers to identify key cost drivers, providing quantitative support for design optimization and cost control (Alshboul et al., 2022a; Banihashemi et al., 2022).

At the data integration level, the deep fusion of BIM with ML is a prominent trend. By extracting geometric parameters and project attributes from BIM models, digital twin–based cost prediction systems have been implemented in large commercial building projects, enabling dynamic cost updates and multi-scenario comparisons (Cheng et al., 2010a).

5.2 Limitations

Despite significant advances in applying ML and AI to construction cost prediction, practical implementation still faces multi-dimensional limitations. As summarized in Table 6, data-related constraints are prominent. Many studies rely on single-project or small-sample datasets—for example, some analyses use only 320 regional samples—resulting in markedly higher MAPE values when models are generalized across regions. Furthermore, data are often concentrated within a single country or region, lacking cross-regional validation. Construction datasets also typically contain approximately 7% missing values and 5% outliers, directly impacting model training efficiency and stability (Gardner et al., 2016; Lee and Yun, 2024; Sadick et al., 2025; Simić et al., 2023).

At the model-method level, deep learning models exhibit a pronounced “black box” problem. Only 23% of construction management professionals can interpret DNN decision logic. Ensemble models, while improving accuracy, incur high computational costs; some hybrid models have over 1,000 parameters, creating hardware and cost barriers for small- and medium-sized enterprises. Feature selection remains largely dependent on expert experience, lacking automated mechanisms, which further limits practical deployment (AlHares and Budayan, 2019; Mahmoodzadeh and Nejati, 2023).

From the application-adaptation perspective, models often show poor cross-project generalization. Performance varies substantially between residential and infrastructure projects, and most models are static, failing to adequately account for temporal dynamics such as material price fluctuations or inflation. Modeling of external uncertainties, including policy or market volatility, is insufficient (Turkyilmaz and Polat, 2025; Cao and Ashuri, 2020).

Evaluation consistency is another limitation. Studies employ different core metrics—RMSE, MAE, MAPE—and adopt varying dataset split ratios, making horizontal comparison of model performance difficult. Collectively, these constraints constitute the main bottlenecks in current applications, highlighting the need for targeted solutions to enhance practical value.

5.3 Future directions

To address current limitations in data availability, model performance, and practical implementation, and to align with the rapid development of artificial intelligence technologies and the intelligent transformation of the construction industry, future research should focus on establishing an integrated “data–model–scenario–deployment” framework. Based on the qualitative analysis of the reviewed literature, Table 7 summarizes potential strategies for overcoming the major challenges identified in current studies.

Small-sample limitations and regional data constraints can be mitigated through data augmentation techniques such as Synthetic Minority Over-sampling Technique (SMOTE) and transfer learning, which help expand datasets and transfer knowledge across different project contexts, thereby supporting applications in specialized projects and data-scarce regions (Zhang et al., 2023). In addition, federated learning frameworks can facilitate cross-institutional data collaboration while preserving data privacy, enabling the integration of multi-source datasets (Mohseni et al., 2025). The fusion of multimodal data derived from Building Information Modeling (BIM), Geographic Information Systems (GIS), and the Internet of Things (IoT) can further promote the establishment of unified data standards and support the deep integration of static and dynamic project information (Shoar et al., 2022).

In terms of model optimization, the integration of Explainable Artificial Intelligence (XAI) techniques with domain-specific knowledge graphs can improve model interpretability by quantifying feature contributions through methods such as SHapley Additive exPlanations (SHAP) values (Shoar et al., 2022). Furthermore, Transformer- and Long Short-Term Memory (LSTM)-based temporal models are capable of capturing dynamic factors, including fluctuations in material prices and market conditions. Hybrid ensemble learning frameworks and Automated Machine Learning (AutoML)-based feature engineering can reduce reliance on expert experience while improving the generalization ability of prediction models across different project types (Ali and Burhan, 2023).

For risk management and uncertainty analysis, Bayesian approaches and probabilistic prediction models can be employed to quantify uncertainty in construction cost estimation. Combined with risk-oriented knowledge graphs, these approaches can support the development of full-lifecycle early warning systems for identifying and predicting potential cost overruns (Zhang et al., 2023). At the deployment stage, lightweight models combined with parameter pruning techniques can reduce computational and hardware requirements, thereby improving practical applicability in real-world engineering environments (Deepa et al., 2025). Moreover, embedding domain knowledge graphs into intelligent prediction systems can enable both cost prediction and optimization-oriented decision support (Salahaldain et al., 2023). Digital twin-driven decision-making systems also have significant potential for enabling dynamic cost monitoring and management throughout the entire project lifecycle (Moghayedi and Windapo, 2023).

Overall, future research should emphasize four key principles: data integration, model transparency, risk quantification, and scenario-oriented applications. Strengthening interdisciplinary collaboration, promoting industry-wide data sharing and standardization, and accelerating the transition from laboratory research to engineering practice will provide comprehensive support for the intelligent development of construction cost management.

6 Conclusion

This study systematically reviewed the research progress of ML and AI in construction cost prediction from 1994 to 2025 by combining scientometric analysis with qualitative synthesis. A total of 138 high-quality journal articles were identified through the Web of Science Core Collection. Using VOSviewer, co-occurrence networks of authors, countries, institutions, and keywords were constructed, revealing the development trajectory, collaboration patterns, and knowledge structure of the field. The research evolution can be characterized by three stages: initial exploration (1994–2008), steady development (2009–2020), and rapid growth (2021–2025). The current research framework centers on ML, with neural networks, ensemble learning, and deep learning as major technical pathways. Key contributors include countries such as China, the United States, and South Korea, as well as institutions like National Taiwan University of Science and Technology and Hanyang University.

Qualitative analysis indicates that key factors influencing construction cost prediction can be categorized into six dimensions: project physical attributes, macroeconomic indicators, resource prices, design parameters, site and environmental conditions, and management and market factors. These factors jointly determine project costs through complex nonlinear relationships and transmission mechanisms. At the technical level, data preprocessing has evolved into a standardized workflow encompassing missing value imputation, outlier detection, normalization, and feature engineering. Model performance evaluation integrates error quantification and generalization validation, providing a unified benchmark for comparing different predictive approaches. Current applications demonstrate substantial improvements in prediction accuracy compared with traditional methods, supporting early-stage estimation, dynamic cost control, and risk warning. Nevertheless, challenges remain, including insufficient data quality, limited interpretability, and weak cross-project generalization.

Future research should focus on four key directions: improving data sharing and quality, integrating multi-source datasets, enhancing model interpretability, and developing dynamic adaptation and risk modeling capabilities. Emerging technologies such as transfer learning, digital twins, explainable AI (XAI), and federated learning offer avenues to overcome these limitations. Strengthening industry collaboration and standardization, including shared datasets, unified evaluation metrics, and deployable engineering tools, will facilitate the transition from laboratory research to practical implementation.

By providing a comprehensive review of current research status and development trends, this study offers a clear knowledge framework and highlights priority directions for future investigation. It also provides theoretical guidance and practical insights for the intelligent transformation of construction cost management. As AI technologies continue to evolve and the digitalization of the construction industry advances, ML and AI applications in cost prediction are expected to become increasingly sophisticated and scalable, supporting lifecycle cost optimization and evidence-based decision-making in construction projects.

References

• AbedY. G.HasanT. M.ZehawiR. N. (2022). Cost prediction for roads construction using machine learning models. Int. Journal Electrical Computer Engineering Systems13 (10), 927–936. 10.32985/ijeces.13.10.8

• Abu-MahfouzE.Al-DahidiS.GharaibehE.AlahmerA. (2025). A novel feature engineering-based hybrid approach for precise construction cost estimation using fuzzy-AHP and artificial neural networks. Int. J. Constr. Manag.25 (15), 1800–1810. 10.1080/15623599.2025.2482207

• AdelK.ElhakeemA.MarzoukM. (2022). Decentralizing construction AI applications using blockchain technology. Expert Syst. Appl.194, 116548. 10.1016/j.eswa.2022.116548

• AhnJ.JiS. H.ParkM.LeeH. S.KimS.SuhS. W. (2014). The attribute impact concept: applications in case-based reasoning and parametric cost estimation. Automation Constr.43, 195–203. 10.1016/j.autcon.2014.03.011

• Al HusaeniD. N. (2023). Bibliometric analysis of research development in sports science with vosviewer. ASEAN J. Phys. Educ. Sport Sci.2 (1), 9–16.

• Al KailaniH.SweisG. J.SammourF.MaaitahW. O.SweisR. J.AlkailaniM. (2025). Predicting construction cost index using fuzzy logic and machine learning in Jordan. Constr. Innov.25 (5), 1479–1500. 10.1108/ci-08-2023-0182

• AlHaresE. F. T.BudayanC. (2019). Estimation at completion simulation using the potential of soft computing models: case study of construction engineering projects. Symmetry11 (2), 190. 10.3390/sym11020190

• AliZ. H.BurhanA. M. (2023). Hybrid machine learning approach for construction cost estimation: an evaluation of extreme gradient boosting model. Asian J. Civ. Eng.24 (7), 2427–2442. 10.1007/s42107-023-00651-z

• AliZ. H.BurhanA. M.KassimM.Al-KhafajiZ. (2022). Developing an integrative data intelligence model for construction cost estimation. Complexity2022 (1), 4285328. 10.1155/2022/4285328

• AlshboulO.ShehadehA.AlmasabhaG.AlmuflihA. S. (2022a). Extreme gradient boosting-based machine learning approach for green building cost prediction. Sustainability14 (11), 6651. 10.3390/su14116651

• AlshboulO.ShehadehA.AlmasabhaG.MamlookR. E. A.AlmuflihA. S. (2022b). Evaluating the impact of external support on green building construction cost: a hybrid mathematical and machine learning prediction approach. Buildings12 (8), 1256. 10.3390/buildings12081256

• AlzaraM.GihadN.AbdouH.SoltanA.HilalA.EhabA. (2025). Macroeconomic-aware forecasting of construction costs in developing countries: using gated recurrent unit and long short-term memory deep learning framework. Plos One20 (10), e0333189. 10.1371/journal.pone.0333189

• AlzubiY.AljaafrehA.KhatatbehA. (2024). Application of machine learning techniques in estimating the construction cost of residential buildings in the Middle East region. Int. J. Constr. Manag.24 (9), 946–958. 10.1080/15623599.2023.2239494

• AnS. H.ParkU. Y.KangK. I.ChoM. Y.ChoH. H. (2007). Application of support vector machines in assessing conceptual cost estimates. J. Comput. Civ. Eng.21 (4), 259–264. 10.1061/(asce)0887-3801(2007)21:4(259)

• AntoniouF.KonstantinidisK. (2025). Client-oriented highway construction cost estimation models using machine learning. Appl. Sci.15 (18), 10237. 10.3390/app151810237

• ArabiatA.Al-BdourH.BisharahM. (2023). Predicting the construction projects time and cost overruns using K-nearest neighbor and artificial neural network: a case study from Jordan. Asian J. Civ. Eng.24 (7), 2405–2414. 10.1007/s42107-023-00649-7

• AydinliS. (2024). Impact of unexpected conditions on construction cost forecasting performance: evidence from Europe. Constr. Manag. Econ.42 (9), 787–801. 10.1080/01446193.2024.2328700

• BanihashemiS.KhaliliS.SheikhkhoshkarM.FazeliA. (2022). Machine learning-integrated 5D BIM informatics: building materials costs data classification and prototype development. Innov. Infrastructure Solutions7 (3), 215. 10.1007/s41062-022-00822-y

• BerawiM. A.SariM.AmiriN. H. A.SusilowatiS. I.UtamiS. R.KulachinskayaA. (2025). Developing a machine learning model to improve the accuracy of owner estimate cost in the capital expenditure procurement process. Int. J. Technol.16 (4), 1179. 10.14716/ijtech.v16i4.7409

• CaoY.AshuriB. (2020). Predicting the volatility of highway construction cost index using long short-term memory. J. Manag. Eng.36 (4), 04020020. 10.1061/(asce)me.1943-5479.0000784

• CaoM. T.ChengM. Y.WuY. W. (2015). Hybrid computational model for forecasting Taiwan construction cost index. J. Constr. Eng. Manag.141 (4), 04014089. 10.1061/(asce)co.1943-7862.0000948

• CaoY.AshuriB.BaekM. (2018). Prediction of unit price bids of resurfacing highway projects through ensemble machine learning. J. Comput. Civ. Eng.32 (5), 04018043. 10.1061/(asce)cp.1943-5487.0000788

• CaoY.ZhangR.QuZ.SkitmoreM.MaX.WangJ. (2026). Forecasting fatal construction accidents using an STL–BiGRU hybrid framework: a multi-scale time series approach. Buildings16 (8), 1539. 10.3390/buildings16081539

• Car-PusicD.PetrusevaS.Zileska PancovskaV.ZafirovskiZ. (2020). Neural network‐based model for predicting preliminary construction cost as part of cost predicting system. Adv. Civil Engineering2020 (1), 8886170. 10.1155/2020/8886170

• ChakrabortyD.ElhegazyH.ElzarkaH.GutierrezL. (2020). A novel construction cost prediction model using hybrid natural and light gradient boosting. Adv. Eng. Inf.46, 101201. 10.1016/j.aei.2020.101201

• ChammoutB.El-AdawayI. H. (2025). Understanding the underlying trends in US construction labor wages: a data-driven mixed-method computational approach. J. Manag. Eng.41 (2), 04024069. 10.1061/jmenea.meeng-6285

• ChenT.GuestrinC. (2016). “Xgboost: a scalable tree boosting system,” in Proceedings of the 22nd Acm Sigkdd International Conference on Knowledge Discovery and Data Mining, 785–794.

• ChenL.TiangS. S.ChongK. S.SharmaA.BerghoutT.LimW. H. (2025a). Predicting construction costs with machine learning: a comparative study on ensemble and linear models. J. Electron. Electromed. Eng. Med. Inf.7 (3), 624–645. 10.35882/jeeemi.v7i3.799

• ChenL.XuC.LimW. H.SharmaA.TiangS. S.ChongK. S.et al (2025b). Transparent and reliable construction cost prediction using advanced machine learning and explainable AI. Eng. Sci. Technol. Int. J.70, 102159. 10.1016/j.jestch.2025.102159

• ChenG.ZhengS.HeX.LiangX.LiaoX. (2025c). Machine learning-based cost estimation models for office buildings. Buildings15 (11), 1802. 10.3390/buildings15111802

• ChengM. Y.HoangN. D. (2014). Interval estimation of construction cost at completion using least squares support vector machine. J. Civ. Eng. Manag.20 (2), 223–236. 10.3846/13923730.2013.801891

• ChengM. Y.KhasaniR. R. (2024). Least square moment balanced machine: a new approach to estimating cost to completion for construction projects. J. Inf. Technol. Constr.29, 503–524. 10.36680/j.itcon.2024.023

• ChengM. Y.TsaiH. C.HsiehW. S. (2009). Web-based conceptual cost estimates for construction projects using evolutionary fuzzy neural inference model. Automation Construction18 (2), 164–172. 10.1016/j.autcon.2008.07.001

• ChengM. Y.PengH. S.WuY. W.ChenT. L. (2010a). Estimate at completion for construction projects using evolutionary support vector machine inference model. Automation Constr.19 (5), 619–629. 10.1016/j.autcon.2010.02.008

• ChengM. Y.TsaiH. C.SudjonoE. (2010b). Conceptual cost estimates using evolutionary fuzzy hybrid neural network for projects in construction industry. Expert Syst. Appl.37 (6), 4224–4231. 10.1016/j.eswa.2009.11.080

• ChengM. Y.HoangN. D.RoyA. F.WuY. W. (2012). A novel time-depended evolutionary fuzzy SVM inference model for estimating construction project at completion. Eng. Appl. Artif. Intell.25 (4), 744–752. 10.1016/j.engappai.2011.09.022

• ChengM. Y.CaoM. T.HeriantoJ. G. (2020). Symbiotic organisms search-optimized deep learning technique for mapping construction cash flow considering complexity of project. Chaos, Solit. and Fractals138, 109869. 10.1016/j.chaos.2020.109869

• ChengM. Y.SholehM. N.PoetraB. A. (2025a). Predicting the construction cost-time tradeoff using optimized hybrid deep learning for risk preference decision making. Int. J. Constr. Manag., 1–27. 10.1080/15623599.2025.2574070

• ChengM. Y.VuQ. T.GosalF. E. (2025b). Hybrid deep learning model for accurate cost and schedule estimation in construction projects using sequential and non-sequential data. Automation Construction170, 105904. 10.1016/j.autcon.2024.105904

• ChouJ. S.LinC. W.PhamA. D.ShaoJ. Y. (2015). Optimized artificial intelligence models for predicting project award price. Automation Construction54, 106–115. 10.1016/j.autcon.2015.02.006

• CoboM. J.López‐HerreraA. G.Herrera‐ViedmaE.HerreraF. (2011). Science mapping software tools: review, analysis, and cooperative study among tools. J. Am. Soc. Information Sci. Technol.62 (7), 1382–1402. 10.1002/asi.21525

• CoffieG. H.CudjoeS. K. F. (2024). Using extreme gradient boosting (XGBoost) machine learning to predict construction cost overruns. Int. J. Constr. Manag.24 (16), 1742–1750. 10.1080/15623599.2023.2289754

• Dang-TrinhN.Duc-ThangP.Nguyen-Ngoc CuongT.Duc-HocT. (2023). Machine learning models for estimating preliminary factory construction cost: case study in Southern Vietnam. Int. J. Constr. Manag.23 (16), 2879–2887. 10.1080/15623599.2022.2106043

• DeepaG.NiranjanaA. J.BaluA. S. (2025). A hybrid machine learning approach for early cost estimation of pile foundations. J. Eng. Des. Technol.23 (1), 306–322. 10.1108/jedt-03-2023-0097

• ElmousalamiH. H. (2020). Comparison of artificial intelligence techniques for project conceptual cost prediction: a case study and comparative analysis. IEEE Transactions Engineering Management68 (1), 183–196. 10.1109/tem.2020.2972078

• FarooqM.ParachaA. T. (2025). Cost prediction of road construction projects in Pakistan using machine learning. Int. J. Constr. Manag.26, 1–29. 10.1080/15623599.2025.2555522

• ForouzeshNejadA. A.ArabikhanF.AheleroffS. (2024). Optimizing project time and cost prediction using a hybrid XGBoost and simulated annealing algorithm. Machines12 (12), 867. 10.3390/machines12120867

• FuC.WangJ.QuZ.SkitmoreM.YiJ.SunZ.et al (2024). Structural equation modeling in technology adoption and use in the construction industry: a scientometric analysis and qualitative review. Sustainability16 (9), 3824. 10.3390/su16093824

• GardnerB. J.GransbergD. D.JeongH. D. (2016). Reducing data-collection efforts for conceptual cost estimating at a highway agency. J. Constr. Eng. Manag.142 (11), 04016057. 10.1061/(asce)co.1943-7862.0001174

• GardnerB. J.GransbergD. D.RuedaJ. A. (2017). Stochastic conceptual cost estimating of highway projects to communicate uncertainty using bootstrap sampling. ASCE-ASME J. Risk Uncertain. Eng. Syst. Part A Civ. Eng.3 (3), 05016002. 10.1061/ajrua6.0000895

• GunaydınH. M.DoganS. Z. (2004). A neural network approach for early cost estimation of structural systems of buildings. Int. J. Proj. Manag.7 (22), 595–602. 10.1016/j.ijproman.2004.04.002

• GurmuA.MiriM. P. (2025). Machine learning regression for estimating the cost range of building projects. Constr. Innov.25 (2), 577–593. 10.1108/ci-08-2022-0197

• HabibO.AbouhamadM.BayoumiA. (2025). Ensemble learning framework for forecasting construction costs. Automation Construction170, 105903. 10.1016/j.autcon.2024.105903

• HeX.LiuR.AnumbaC. J. (2021). Data-driven insights on the knowledge gaps of conceptual cost estimation modeling. J. Constr. Eng. Manag.147 (2), 04020165. 10.1061/(asce)co.1943-7862.0001963

• HelalyH.El-RayesK.IgnacioE. J.JoanH. J. (2025). Comparison of machine-learning algorithms for estimating cost of conventional and accelerated bridge construction methods during early design phase. J. Constr. Eng. Manag.151 (3), 04025004. 10.1061/jcemd4.coeng-15934

• HosseiniM. R.MartekI.ZavadskasE. K.AibinuA. A.ArashpourM.ChilesheN. (2018). Critical evaluation of off-site construction research: a scientometric analysis. Automation Construction87, 235–247. 10.1016/j.autcon.2017.12.002

• JaafariA.PazhouhanI.BettingerP. (2021). Machine learning modeling of forest road construction costs. Forests12 (9), 1169. 10.3390/f12091169

• JafariM.MousaviE. (2026). A critical review of the data-driven cost estimation approach in construction research. J. Constr. Eng. Manag.152 (2), 03125013. 10.1061/jcemd4.coeng-16840

• JafaryP.ShojaeiD.RajabifardA.NgoT. (2025). AI-augmented construction cost estimation: an ensemble Natural Language Processing (NLP) model to align quantity take-offs with cost indexes. Int. J. Constr. Manag.26, 1–19. 10.1080/15623599.2025.2558070

• JinR.ZouP. X.PiroozfarP.WoodH.YangY.YanL.et al (2019). A science mapping approach based review of construction safety research. Saf. Science113, 285–297. 10.1016/j.ssci.2018.12.006

• JuszczykM. (2019). On the search of models for early cost estimates of bridges: an SVM-based approach. Buildings10 (1), 2. 10.3390/buildings10010002

• JuszczykM.LeśniakA. (2019). Modelling construction site cost index based on neural network ensembles. Symmetry11 (3), 411. 10.3390/sym11030411

• KaracaI.GransbergD. D.JeongH. D. (2020). Improving the accuracy of early cost estimates on transportation infrastructure projects. J. Manag. Eng.36 (5), 04020063. 10.1061/(asce)me.1943-5479.0000819

• KimJ. S. (2025a). AI-powered forecasting of environmental impacts and construction costs to enhance project management in highway projects. Buildings15 (14), 2546. 10.3390/buildings15142546

• KimJ. S. (2025b). Advancing sustainable road construction with multiple regression analysis, regression tree models, and case-based reasoning for environmental load and cost estimation. Buildings15 (12), 2083. 10.3390/buildings15122083

• KimS. Y.ChoiJ. W.KimG. H.KangK. I. (2005). Comparing cost prediction methods for apartment housing projects: CBR versus ANN. J. Asian Archit. Build. Eng.4 (1), 113–120. 10.3130/jaabe.4.113

• LarocqueR.BouleA. M.CappartQ. (2025). Estimating road construction costs with explainable machine learning. Inf. J. Appl. Anal.55 (2), 137–153. 10.1287/inte.2023.0027

• LarsenP.Von InsM. (2010). The rate of growth in scientific publication and the decline in coverage provided by Science Citation Index. Scientometrics84 (3), 575–603. 10.1007/s11192-010-0202-z

• LeeH.YunS. (2024). Strategies for imputing missing values and removing outliers in the dataset for machine learning-based construction cost prediction. Buildings14 (4), 933. 10.3390/buildings14040933

• LiuZ.LinJ. (2025). Combinatorial machine learning approaches for high-rise building cost prediction and their interpretability analysis. J. Asian Archit. Build. Eng.24 (4), 2817–2828. 10.1080/13467581.2024.2378001

• MahdavianA.ShojaeiA.SalemM.YuanJ. S.OloufaA. A. (2021). Data-driven predictive modeling of highway construction cost items. J. Constr. Eng. Manag.147 (3), 04020180. 10.1061/(asce)co.1943-7862.0001991

• MahmoodzadehA.NejatiH. R. (2023). An optimized equation based on the gene expression programming method for estimating tunnel construction costs considering a variety of variables and indexes. Appl. Soft Comput.147, 110749. 10.1016/j.asoc.2023.110749

• MahmoodzadehA.MohammadiM.DaraeiA.Farid Hama AliH.Ismail AbdullahA.Kameran Al-SalihiN. (2021). Forecasting tunnel geology, construction time and costs using machine learning methods. Neural Comput. Appl.33 (1), 321–348. 10.1007/s00521-020-05006-2

• MahmoodzadehA.NejatiH. R.MohammadiM. (2022). Optimized machine learning modelling for predicting the construction cost and duration of tunnelling projects. Automation Constr.139, 104305. 10.1016/j.autcon.2022.104305

• MahmoodzadehA.NejatiH. R.FlaihL. R.Hashim IbrahimH.AleniziF. A.AlassafY. (2024). A rigorous examination of twelve cutting-edge machine-learning techniques for predicting time and cost in tunneling projects. J. Constr. Eng. Manag.150 (12), 04024176. 10.1061/jcemd4.coeng-15070

• MarzoukM.ElhakeemA.AdelK. (2024). Artificial neural networks applications in construction and building engineering (1991–2021): science mapping and visualization. Appl. Soft Comput.152, 111174. 10.1016/j.asoc.2023.111174

• MeharieM. G.MengeshaW. J.GariyZ. A.MutukuR. N. (2022). Application of stacking ensemble machine learning algorithm in predicting the cost of highway construction projects. Eng. Constr. Archit. Manag.29 (7), 2836–2853. 10.1108/ecam-02-2020-0128

• MengC.QuD.DuanX. (2024). Cost estimation of metro construction projects using interpretable machine learning. J. Comput. Civ. Eng.38 (6), 04024038. 10.1061/jccee5.cpeng-6018

• MhadyA. A.GurgunA. P.BudayanC.KocK. (2025). New hybrid models integrating the firefly optimization algorithm with the artificial neural networks and adaptive neuro-fuzzy inference systems to improve estimation at completion. J. Constr. Eng. Manag.151 (11), 04025160. 10.1061/jcemd4.coeng-16487

• MoghayediA.WindapoA. (2023). Evaluating the impact of uncertainty events on the cost of linear infrastructure projects. Proc. Institution Civ. Engineers-Engineering Sustain.176(6), 323–334. 10.1680/jensu.21.00061

• MohamedB.MoselhiO. (2022). Conceptual estimation of construction duration and cost of public highway projects. J. Inf. Technol. Constr.27 (29), 595–618. 10.36680/j.itcon.2022.029

• MoherD.LiberatiA.TetzlaffJ.AltmanD. G.Prisma Group (2010). Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. Int. Journal Surgery8 (5), 336–341. 10.1016/j.ijsu.2010.02.007

• MohseniM.KamalE. M.Akhavan TabassiA. (2025). Cost overrun prediction model for petrochemical projects in Iran. J. Constr. Dev. Ctries.30 (s2), 89–117. 10.21315/jcdc.2025.30.s2.5

• MolnarC.KönigG.HerbingerJ.FreieslebenT.DandlS.ScholbeckC. A.et al (2020). “General pitfalls of model-agnostic interpretation methods for machine learning models,” in International Workshop on Extending Explainable AI Beyond Deep Models and Classifiers (Cham: Springer International Publishing), 39–68.

• MostofiF.ToğanV.TokdemirO. B.ArayiciY. (2025). A cost estimation recommendation system for improved contingency management in construction projects. Neural Comput. Appl.37 (5), 3521–3538. 10.1007/s00521-024-10740-y

• NarbaevT.HazirÖ.KhamitovaB.TalgatS. (2024). A machine learning study to improve the reliability of project cost estimates. Int. J. Prod. Res.62 (12), 4372–4388. 10.1080/00207543.2023.2262051

• NewmanM. E. (2004). Coauthorship networks and patterns of scientific collaboration. Proc. National Academy Sciences101 (Suppl. l_1), 5200–5205. 10.1073/pnas.0307545100

• NguyenT. T.NguyenD. D.NguyenS. D.PrakashI.Van TranP.PhamB. T. (2022). Forecasting construction price index using artificial intelligence models: support vector machines and radial basis function neural network. J. Sci. Transp. Technol., 2 (4), 9–19. 10.58845/jstt.utt.2022.en.2.4.9-19

• NiaziA.DaiJ. S.BalabaniS.SeneviratneL. (2006). Product Cost Estimation: Technique Classification and Methodology Review.

• NindartinA.ParkS. J.LeeK. T.KimJ. H.RostiyantiS. F. (2025). Prediction of cost contingency in construction projects by introducing machine learning algorithms. J. Civ. Eng. Manag.31 (8), 860–880. 10.3846/jcem.2025.24913

• OztekinA.MastersonR. R. (2018). A predictive analytic approach to determine construction cost estimates. Int. J. Operational Res.32 (2), 169–200. 10.1504/ijor.2018.10013121

• PadalaS. S.GoyalA. (2025). Early stage cost prediction model for Indian building construction projects using artificial neural networks. J. Financial Manag. Prop. Constr.30 (3), 377–397. 10.1108/jfmpc-12-2023-0085

• ParkD.YunS. (2023). Construction cost prediction using deep learning with BIM properties in the schematic design phase. Appl. Sci.13 (12), 7207. 10.3390/app13127207

• ParkU.KangY.LeeH.YunS. (2022). A stacking heterogeneous ensemble learning method for the prediction of building construction project costs. Appl. Sciences12 (19), 9729. 10.3390/app12199729

• ParkS. J.NourN.LeeK. Y.KimJ. H. (2025). Prediction of sewage pipeline construction duration by introducing machine learning and deep learning approaches. J. Civ. Eng. Manag.31 (7), 687–709. 10.3846/jcem.2025.23472

• PeškoI.MučenskiV.ŠešlijaM.RadovićN.VujkovA.BibićD.et al (2017). Estimation of costs and durations of construction of urban roads using ANN and SVM. Complexity2017 (1), 2450370. 10.1155/2017/2450370

• PessoaA.SousaG.Furtado MauésL. M.Campos AlvarengaF.SantosD. D. G. (2021). Cost forecasting of public construction projects using multilayer perceptron artificial neural networks: a case study. Ing. Investig.41 (3), e87737. 10.15446/ing.investig.v41n3.87737

• PhamT. Q. D.Le-HongT.TranX. V. (2023). Efficient estimation and optimization of building costs using machine learning. Int. J. Constr. Manag.23 (5), 909–921. 10.1080/15623599.2021.1943630

• RafieiM. H.AdeliH. (2018). Novel machine-learning model for estimating construction costs considering economic variables and indexes. J. Construction Engineering Management144 (12), 04018106. 10.1061/(asce)co.1943-7862.0001570

• RashadA.HegazyH.ZhangJ.MahdiI.Abdel-RasheedI.EbidA. (2025). Developing preliminary cost estimates for foundation systems of high-rise buildings. Int. J. Constr. Manag.25 (6), 682–698. 10.1080/15623599.2024.2352180

• RokickiB.OstaszewskiK. (2022). Actuarial credibility approach in adjusting initial cost estimates of transport infrastructure projects. Sustainability14 (20), 13371. 10.3390/su142013371

• SadickA. M.GurmuA.GunarathnaC. (2025). Artificial intelligence-based pre-conception stage construction budget decision-making model and tool for residential buildings. Eng. Constr. Archit. Manag.32 (7), 4552–4580. 10.1108/ecam-11-2023-1108

• SaeidlouS.GhadiminiaN. (2024). A construction cost estimation framework using DNN and validation unit. Build. Res. and Inf.52 (1-2), 38–48. 10.1080/09613218.2023.2196388

• SalahaldainZ.NaimiS.AlsultaniR. (2023). Estimation and analysis of building costs using artificial intelligence support vector machine. Math. Model. Eng. Problems10 (2), 405–411. 10.18280/mmep.100203

• Sanni-AnibireM. O.ZinR. M.OlatunjiS. O. (2022). “Developing a preliminary cost estimation model for tall buildings based on machine learning,” in Big Data and Information Theory (London: Routledge), 94–102.

• SeruggaJ. (2025). AI-Assisted game theory approaches to bid pricing under uncertainty in construction. AppliedMath5 (2), 39. 10.3390/appliedmath5020039

• ShamimM. M. I.HamidA. B. B. A.NyamasvisvaT. E.RafiN. S. B. (2025). Advancement of artificial intelligence in cost estimation for project management success: a systematic review of machine learning, deep learning, regression, and hybrid models. Modelling6 (2), 35. 10.3390/modelling6020035

• SharmaV.ZakiM.JhaK. N.KrishnanN. A. (2022). Machine learning-aided cost prediction and optimization in construction operations. Eng. Constr. Archit. Manag.29 (3), 1241–1257. 10.1108/ECAM-10-2020-0778

• Shartooh SharqiS.BhattaraiA. (2021). Evaluation of several machine learning models for field canal improvement project cost prediction. Complexity2021 (1), 8324272. 10.1155/2021/8324272

• ShenY. (2025). Using regression-based machine learning model to estimate constructions cost. Int. J. Inf. Commun. Technol.26 (17), 72–92. 10.1504/ijict.2025.146673

• ShinY. (2015). Application of boosting regression trees to preliminary cost estimation in building construction projects. Comput. Intelligence Neuroscience2015 (1), 149702. 10.1155/2015/149702

• ShoarS.ChilesheN.EdwardsJ. D. (2022). Machine learning-aided engineering services’ cost overruns prediction in high-rise residential building projects: application of random forest regression. J. Build. Eng.50, 104102. 10.1016/j.jobe.2022.104102

• SimićN.IvaniševićN.NedeljkovićĐ.SenićA.StojadinovićZ.IvanovićM. (2023). Early highway construction cost estimation: selection of key cost drivers. Sustainability15 (6), 5584. 10.3390/su15065584

• SitthikankunS.RinchumphuD.BuachartC.PacharawongsakdaE. (2021). Construction cost estimation for government building using artificial neural network. Int. Trans. J. Eng. Manag. Appl. Sci. Technol.12, 1–12. 10.14456/ITJEMAST.2021.112

• SohrabiH.NoorzaiE. (2024). Risk-supported case-based reasoning approach for cost overrun estimation of water-related projects using machine learning. Eng. Constr. Archit. Manag.31 (2), 544–570. 10.1108/ecam-05-2022-0450

• SunejaN.ShahJ. P.ShahZ. H.HoliaM. S. (2021). A neural network approach to design reality oriented cost estimate model for infrastructure projects. Reliab. Theory and Appl.16 (60), 254–263.

• TakefujiY. (2021). Converting detailed estimates to primary estimates with data augmentation. Adv. Eng. Inf.49, 101354. 10.1016/j.aei.2021.101354

• TangS.ShideK. (2025). “Enhancing deep learning models for early-stage construction cost estimation through optimized feature engineering,” in 30th International Conference on Computer-Aided Architectural Design Research in Asia, CAADRIA 2025 (Hong Kong: Association for Computer-Aided Architectural Design Research in Asia), 101–110.

• TurkyilmazA. H.PolatG. (2024). Risk-based completion cost overrun ratio estimation in construction projects using machine learning classification algorithms: a case study. Buildings14 (11), 3541. 10.3390/buildings14113541

• TurkyilmazA. H.PolatG. (2025). Risk-based completion cost prediction approach in construction projects utilizing machine learning. J. Inf. Technol. Constr.30, 375–396. 10.36680/j.itcon.2025.016

• UgurL. (2017). A neuro-adaptive learning (NAL) approach about costs of residential buildings. Acta Phys. Pol. A132 (3), 585–587. 10.12693/aphyspola.132.585

• UgurL. O.KanitR.ErdalH.NamliE.ErdalH. I.BaykanU. N.et al (2019). Enhanced predictive models for construction costs: a case study of Turkish mass housing sector. Comput. Econ.53 (4), 1403–1419. 10.1007/s10614-018-9814-9

• UysalF.SonmezR. (2023). Bootstrap aggregated case-based reasoning method for conceptual cost estimation. Buildings13 (3), 651. 10.3390/buildings13030651

• Van EckN.WaltmanL. (2010). Software survey: vosviewer, a computer program for bibliometric mapping. Scientometrics84 (2), 523–538. 10.1007/s11192-009-0146-3

• WangJ.AshuriB. (2017). Predicting ENR construction cost index using machine-learning algorithms. Int. Journal Construction Education Research13 (1), 47–63. 10.1080/15578771.2016.1235063

• WangB.YuanJ.GhafoorK. Z. (2021). Research on construction cost estimation based on artificial intelligence technology. Scalable Comput. Pract. Exp.22 (2), 93–104. 10.12694/scpe.v22i2.1868

• WangY. R.YuC. Y.ChanH. H. (2012). Predicting construction cost and schedule success using artificial neural networks ensemble and support vector machines classification models. Int. J. Proj. Manag.30 (4), 470–478. 10.1016/j.ijproman.2011.09.002

• WangJ.ChenJ.HuY. (2022a). A science mapping approach based review of model predictive control for smart building operation management. J. Civ. Eng. Manag.28 (8), 661–679. 10.3846/jcem.2022.17566

• WangR.AsghariV.CheungC. M.HsuS. C.LeeC. J. (2022b). Assessing effects of economic factors on construction cost estimation using deep neural networks. Automation Constr.134, 104080. 10.1016/j.autcon.2021.104080

• WangJ.LiM.SkitmoreM.ChenJ. (2024). Predicting construction company insolvent failure: a scientometric analysis and qualitative review of research trends. Sustainability16 (6), 2290. 10.3390/su16062290

• WangJ.QuZ.SkitmoreM. (2025a). Predicting construction cost index using a VMD-GRU framework with the MHSA mechanism. J. Constr. Eng. Manag.152 (1), 04025236. 10.1061/jcemd4.coeng-17141

• WangJ.GuQ.ChilesheN. (2025b). Construction cost prediction using a fused loss–enhanced hybrid GCN–GAT–MLP approach. Eng. Constr. Archit. Manag., 1–24. 10.1108/ecam-07-2025-1081

• WangJ.QuZ.LeeC. Y.SkitmoreM. (2025c). Highway construction cost index forecasting: a hybrid VMD–LSTM–GRU method. Constr. Manag. Econ.43 (10), 849–863. 10.1080/01446193.2025.2525871

• WangR.SallehH.LiK.Abdul-SamadZ. (2025d). A conceptual cost estimation model for building construction projects by hybrid back-propagation neural network and dung beetle optimizer algorithm. J. Asian Archit. Build. Eng.24 (6), 5520–5542. 10.1080/13467581.2024.2445596

• WangR.SallehH.LyuJ.Abdul-SamadZ.RadzuanN. F. M.WenK. C. (2025e). Application and prospect of machine learning techniques in cost estimation of building projects. Eng. Constr. Archit. Manag.32 (12), 8445–8471. 10.1108/ecam-05-2024-0595

• WangJ.QuZ.ZhangR.SkitmoreM. (2026a). Predicting construction safety accidents using VMD-BiGRU architecture with improved snake optimization algorithm. Eng. Constr. Archit. Manag., 1–27. 10.1108/ECAM-10-2025-1596

• WangJ.QuZ.ZhangR.ChilesheN.JiZ. (2026b). Forecasting the architecture billings index using a variational mode decomposition–bidirectional gated recurrent unit framework based on grey wolf optimizer. Eng. Constr. Archit. Manag., 1–25. 10.1108/ECAM-12-2025-1983

• WuZ.LiuM.MaG.JiangS. (2025a). A hybrid forecasting model to improve cost prediction accuracy in green building projects with machine learning. Eng. Constr. Archit. Manag.33 (4), 3375–3402. 10.1108/ECAM-09-2024-1247

• WuZ.LiuM.MaG. (2025b). A machine learning-based two-stage integrated framework for cost reasonableness prediction of green building projects. J. Build. Eng.100, 111733. 10.1016/j.jobe.2024.111733

• XuH.ChangR.PanM.LiH.LiuS.WebberR. J.et al (2022). Application of artificial neural networks in construction management: a scientometric review. Buildings12 (7), 952. 10.3390/buildings12070952

• YalçınG.BayramS.ÇıtakoğluH. (2024). Evaluation of earned value management-based cost estimation via machine learning. Buildings14 (12), 3772. 10.3390/buildings14123772

• YamusaM. A.LawalH. S. I.AbdulrahmanR. S.SalisuA. S.SakaA.AbubakarM.et al (2025). A machine-learning model for estimating construction renovation costs. Int. J. Build. Pathology Adapt.44 (3), 757–772. 10.1108/IJBPA-10-2024-0213

• YevuS. K.AnnT. W.DarkoA. (2021). Digitalization of construction supply chain and procurement in the built environment: emerging technologies and opportunities for sustainable processes. J. Clean. Prod.322, 129093. 10.1016/j.jclepro.2021.129093

• YiZ.LuoX. (2024). Construction cost estimation model and dynamic management control analysis based on artificial intelligence. Iran. J. Sci. Technol. Trans. Civ. Eng.48 (1), 577–588. 10.1007/s40996-023-01173-z

• YuS. (2025). Data processing of whole process cost management in building engineering based on AI. Intell. Decis. Technol.19 (3), 1464–1476. 10.1177/18724981251317660

• YunS. (2022). Predictive performance of building construction estimation: an analysis based on ANN model. J. Syst. Manag. Sci.12 (2), 331–341. 10.33168/JSMS.2022.0216

• ZangH.LiM.JinZ.HuangJ. (2025). Unveiling construction accident causation: a scientometric analysis and qualitative review of research trends. Front. Built Environ.11, 1602297. 10.3389/fbuil.2025.1602297

• ZengZ.GaoY. (2024). Cost control management of construction projects based on fuzzy logic and auction theory. Ieee Access12, 130292–130304. 10.1109/access.2024.3438291

• ZhaiZ.ShanM.DarkoA.LeY. (2020). Visualizing the knowledge domain of project governance: a scientometric review. Adv. Civ. Eng.2020 (1), 6813043. 10.1155/2020/6813043

• ZhangY. (2024). Research on the application of artificial intelligence-based cost estimation and cost control methods in green buildings. Scalable Comput. Pract. Exp.25 (6), 4772–4779. 10.12694/scpe.v25i6.3293

• ZhangY.Minchin JrR. E.FloodI.RiesR. J. (2023). Preliminary cost estimation of highway projects using statistical learning methods. J. Constr. Eng. Manag.149 (5), 04023026. 10.1061/jcemd4.coeng-12773

• ZhangJ.WangJ.ZangH.MaN.SkitmoreM.QuZ.et al (2024). The application of machine learning and deep learning in intelligent transportation: a scientometric analysis and qualitative review of research trends. Sustainability16 (14), 5879. 10.3390/su16145879

• ZhengZ.ZhouL.WuH.ZhouL. (2023). Construction cost prediction system based on random forest optimized by the bird swarm algorithm. Math. Biosci. Eng.20 (8), 15044–15074. 10.3934/mbe.2023674

• ZhouJ. Q.LiuQ. M.MaC. X.LiD. (2024). Cost prediction of tunnel construction based on interpretative structural model and stacked sparse autoencoder. Eng. Lett.32 (10), 1966–1980.

• ZhouL.LiuC.SongX. (2025). Identification of core sub-team on scientific collaboration networks with Shapley method. PeerJ Comput. Sci.11, e3048. 10.7717/peerj-cs.3048