Bridging policy and practice: understanding civil engineers’ adoption of biophilic development in Thailand using the Theory of Planned Behavior and explainable AI Open Access

Urban areas across the globe are grappling with unprecedented challenges driven by rapid urbanization, climate change and environmental degradation. These pressures threaten human well-being and underscore the urgent need for sustainable, resilient urban development models (Africa et al., 2019). Biophilic development, which integrates natural elements into the built environment, has emerged as a promising strategy to address these complex urban issues. By leveraging the innate human connection to nature, biophilic design mitigates urban heat island effects, improves air quality, enhances biodiversity and supports healthier, more livable cities (Newman et al., 2012; Africa et al., 2019; Abdullah et al., 2024).

Despite these advancements, the application of biophilic design remains fragmented, particularly within large-scale infrastructure projects traditionally led by civil engineers. While architects and landscape architects have increasingly championed biophilic approaches, civil engineers often prioritize conventional technical and economic considerations, marginalizing ecological integration. Barriers such as limited knowledge, perceived cost implications, and the absence of supportive regulatory frameworks have contributed to the slow uptake of biophilic principles in Thailand’s infrastructure sector (Aruninta and Matsushima, 2022; Debrah et al., 2020).

To investigate the factors influencing civil engineers’ adoption of biophilic development, this study employs a multidisciplinary research approach that combines statistical analysis, exploratory data analysis (EDA) and explainable artificial intelligence (XAI) techniques. Random forest (RF) models are used to predict decision patterns, while Shapley Additive Explanations (SHAP) provide interpretability of model outcomes (Love et al., 2023; Naser, 2021; Feng et al., 2021; Kashem et al., 2024). This integrated approach enables both robust predictive modeling and a transparent understanding of behavioral drivers in engineering decision-making.

The objective of this study is to examine key behavioral factors that influence Thai civil engineers’ intention to adopt biophilic development (IntBD) principles in infrastructure projects. Based on the TPB, the research focuses on the roles of attitudes, subjective norms, perceived behavioral control (PBC) and knowledge in shaping adoption intentions, while also identifying barriers and enabling conditions for integrating biophilic concepts into practice. Unlike previous studies that focus mainly on architecture or policy, this study highlights the engineering decision-making process using both behavioral theory and explainable machine learning. The findings of this study are intended to deliver practical benefits for a range of stakeholders. For civil engineers, the insights will offer guidance on integrating biophilic principles into infrastructure projects, enhancing both environmental performance and project sustainability. Governmental agencies, such as the Ministry of Transport and Engineering Councils, can use the findings to inform policy development, regulatory frameworks and capacity-building initiatives that encourage biophilic adoption. Urban planners and policymakers will gain evidence-based recommendations to incorporate biophilic strategies into urban development plans, advancing sustainable and resilient city-making efforts. This study is among the first to integrate the Theory of Planned Behavior (TPB) with explainable machine learning techniques to examine civil engineers’ intention to adopt biophilic development, offering a novel approach to understanding behavioral drivers in sustainable infrastructure design.

Globally, initiatives such as Singapore’s expansive greenway networks and New York City’s High Line project demonstrate how biophilic approaches can revitalize urban landscapes, offering ecological, social and economic benefits (Sini, 2020; Lang and Rothenberg, 2017; An et al., 2020). These projects align closely with multiple Sustainable Development Goals (SDGs), particularly SDG 11 (Sustainable Cities and Communities), SDG 13 (Climate Action) and SDG 15 (Life on Land), positioning biophilic development as a cornerstone of sustainable urban futures (Söderlund, 2019; United Nations, 2023). Thailand has begun integrating biophilic principles into certain sectors, notably through the greening of educational campuses (Abdul Malek et al., 2024; Atthakorn, 2022) and the rise of sustainable residential developments (Rauf et al., 2024; Paramita et al., 2022). Moreover, nature-based Solutions (NbS) have been adopted in urban planning initiatives addressing flood resilience and stormwater management (Vojinovic et al., 2021; Irvine et al., 2023; Ahmed et al., 2024).

Civil engineers are pivotal to the realization of infrastructure projects that shape urban form and environmental outcomes. Yet, little is known about the behavioral factors influencing their willingness to incorporate biophilic design into practice. Most existing research focuses on design professionals outside the engineering domain, overlooking the distinct roles, responsibilities and decision-making contexts of civil engineers (Alsabban and Bettaieb, 2022; Mcgee et al., 2019) Although there is a tendency in the literature to present successful biophilic initiatives, such as Singapore’s park networks, in a largely descriptive manner (Sini, 2020), there has been limited critical examination of the contextual differences between countries. While Singapore’s success is often highlighted, its highly centralized planning and strong regulatory support differ significantly from Thailand’s more decentralized and fragmented approach, making direct application of its strategies more challenging. In particular, Thailand differs substantially from Singapore in terms of policy frameworks, governance structures and urban development trajectories. These contextual differences suggest that the success of biophilic initiatives in Singapore may not be directly replicable in Thailand. Despite global and local interest in biophilic development, little is known about how civil engineers in Thailand form intentions to adopt such principles. Existing literature focuses mainly on architectural design or policy, often overlooking behavioral and contextual barriers specific to the engineering field. Addressing this gap, the present study investigates the key psychological and contextual factors influencing Thai civil engineers’ IntBD principles, using the TPB as the guiding framework (Ajzen, 1991) (Figure 1).

This study conducted an online cross-sectional survey targeting Thai civil engineers who hold a professional license and possessing at least one year of experience in designing and implementing infrastructure development and management projects. A convenience sampling method was employed to ensure accessibility to participants. As no prior studies or pilot data exist on engineers' knowledge, attitudes and behaviors toward biophilic development, the sample size was calculated using a standard formula for estimating proportions in cross-sectional studies with an unknown population size (Charan and Biswas, 2013). A prevalence (p) of 0.5 was adopted, representing maximum variability and ensuring the largest possible sample size as a conservative estimate. The calculation assumed a 95% confidence level and a 5% margin of error, resulting in a required sample size of 385 participants.

Data were collected using an online questionnaire, developed through a literature review and structured according to the TPB framework (Ajzen, 1991). The questionnaire consisted of three main sections. The general information section included 10 items covering demographic details such as age, educational level, work experience, organizational position and behavioral experience related to the implementation of biophilic development. The attitudes toward biophilic development implementation section comprised 23 items, exploring participants’ perceptions of the benefits and barriers associated with integrating biophilic development into infrastructure projects. Responses were measured using a 5-point Likert scale (5 = strongly agree, 4 = agree, 3 = neutral, 2 = disagree, 1 = strongly disagree) as it provides a structured and widely accepted method for capturing complex attitudes and perceptions, allowing meaningful analysis even when the data are ordinal in nature (Sullivan and Artino, 2013). The final section, knowledge about biophilic development, included 5 multiple-choice questions assessing participants’ understanding of biophilic development.

The questionnaire was tested to ensure accuracy and reliability. Content validity was evaluated by three experts: one engineering specialist, one academic engineer and one biophilic expert. The Index of Item-Objective Congruence (IOC) was calculated, with all items achieving an IOC score above 0.67, indicating acceptable content validity. For reliability testing, the revised questionnaire was pilot tested with 15 participants who were not part of the main study sample. Cronbach’s alpha coefficient was calculated, yielding values of 0.87 for knowledge and 0.86 for attitudes, indicating a high level of internal consistency and reliability of the instrument.

An EDA was performed on the collected data to uncover patterns, relationships and critical characteristics associated with biophilic development adoption. Data visualization was conducted using Microsoft Power BI, employing various analytical tools to explore the distribution and interactions of key features. The analysis began with target variables exploration through visual representations such as donut charts for categorical attributes, including demographic characteristics. Correlation analysis was then conducted using scatter plots to examine the relationships between key behavioral factors, including attitude, subjective norms and PBC. To further explore variable interactions, pair plots were generated, highlighting the influence of knowledge-related variables on the adoption of biophilic development principles. This multi-layered analytical approach provided a comprehensive understanding of how key variables interact and contribute to the adoption process.

Binary logistic regression was used to examine the relationships between attitudes, subjective norms, PBC and knowledge, and IntBD, with analysis conducted using SPSS software. This method was selected for its suitability in modeling binary outcomes, and all statistical tests were performed at a significance level of p < 0.05. Key outputs included odds ratios (OR) and adjusted odds ratios (aOR) with 95% confidence intervals (CI), indicating the strength and direction of associations between predictors and IntBD. Univariate analysis was used to assess each variable individually, while multivariate analysis (binary logistic regression) provided a comprehensive understanding of how these factors interact to influence adoption intentions (Sperandei, 2014). This combined approach ensured both statistical rigor and practical insight for informing policy and intervention strategies.

To gain deeper insights into the dataset, a machine learning approach using the RF algorithm was used. Two RF models were developed using the Python scikit-learn library, focusing on predicting biophilic development awareness (AweBD) and adoption (IntBD). Hyperparameter tuning was conducted using both grid search and random search techniques (Russell and Norvig, 2021) to optimize the models' predictive performance. Key parameters such as the number of trees (n_estimators) and maximum tree depth (max_depth) were fine-tuned to enhance model accuracy and reliability. Following the automated tuning process, further manual adjustments were made to refine the models’ performance. The models’ effectiveness was evaluated using various performance metrics, including accuracy, precision, recall, F1-score, confusion matrix, receiver operating characteristic (ROC) curve and precision-recall curve. Additionally, the models’ feature importance scores were analyzed, highlighting the top seven most influential features driving the predictions. This comprehensive evaluation provided valuable insights into the key factors influencing engineers’ awareness and IntBD principles.

To complement the statistical findings from binary logistic regression, SHAP analysis was used to enhance the interpretability of the machine learning model and provide deeper insights into the relative importance of each predictor. While logistic regression revealed statistically significant associations between attitudes, subjective norms, PBC and knowledge with IntBD, SHAP enabled instance-level, model-agnostic interpretation of how each variable influenced predictions made by the random forest model.

SHAP, a widely used method in explainable artificial intelligence (XAI), assigns each feature a contribution value (SHAP value) for every prediction. In this study, SHAP values were calculated using the SHAP Python library based on the same dataset used to train the model. These values offered a transparent explanation of how each feature increased or decreased the predicted probability of IntBD on a case-by-case basis.

The analysis began with SHAP summary plots to visualize the overall impact of each feature across all cases. Additionally, SHAP force plots were used to illustrate how individual features pushed predictions higher or lower, helping make the model’s reasoning more accessible.

By integrating SHAP analysis with traditional statistical methods, this study not only identified key predictors of IntBD but also clarified their influence at both group and individual levels, supporting evidence-based policy and design strategies aimed at encouraging biophilic development.

The results of the survey show a significant gap between awareness and intention regarding biophilic development. The left chart displayed in Figure 2(a) responses to AweBD (do you know biophilic development?), where only 25.1% of respondents indicated awareness of biophilic development principles, while the majority, 74.9%, were unaware. This finding underscores a considerable knowledge gap among engineers about biophilic development, emphasizing the need for targeted educational initiatives to increase awareness of its principles and benefits. In contrast, responses to the right chart displayed in Figure 2(b) to IntBD (do you intend to adopt biophilic development?). Despite the low awareness indicated in AweBD, a significant 68.2% of respondents expressed an IntBD principle in their work, with only 31.8% indicating no intention to adopt. This suggests that even with limited knowledge of biophilic development, there is a strong willingness among engineers to embrace sustainable practices when provided with the necessary support and guidance.

The demographic characteristics of the survey respondents are displayed in Figure 3. The age distribution displayed in Figure 3(a) shows that nearly half (46.67%) of respondents fall within the 40–49 age group, making it the largest segment. The 30–39 age group accounts for 24.62%, while 16.41% are aged 50–59. Younger respondents, under the age of 30, represent 9.74%, and only 2.56% are aged 60 or above. In terms of education displayed in Figure 3(b), the majority of respondents (51.79%) hold an undergraduate degree, while 46.15% have completed a master’s or Ph.D. Only 2.05% have a high school-level education. Regarding professional titles displayed in Figure 3(c), the largest group (57.95%) consists of Assistant Engineers, followed by Engineers at 28.72% and Senior Engineers at 13.33%. Finally, Figure 3(d) displays the distribution of years of experience. The largest proportion of respondents (44.1%) has 20–29 years of experience, while 27.69% have fewer than 10 years, and 17.44% have 10–19 years. Only 10.77% of respondents reported having more than 30 years of experience.

Two initial analyses were conducted by cross-analyzing the data displayed in Figures 2 and 3. The demographic characteristics of respondents who were aware of biophilic development (AweBD) are shown in Figure 4, while those who expressed an IntBD are displayed in Figure 5. The values highlighted in these figures follow a consistent pattern, illustrating the differences between respondents with awareness or intention and those without.

For individuals who are AweBD, Figure 4(a) highlights that those aged 50–59 have the highest level of awareness compared to other age groups, followed by individuals under 30. Figure 4(b) shows that individuals with a Master’s or Ph.D. degree have a higher awareness ratio than those with a Bachelor’s degree. However, while the chart indicates the highest ratio among individuals with only a high school education, this data was omitted due to the small sample size. Figure 4(c) illustrates that senior engineers exhibit the highest awareness, followed by engineers and assistant engineers. Figure 4(d) indicates that individuals with more than 30 years of experience have the highest awareness ratio. Overall, the analysis suggests that older individuals with extensive experience and higher professional titles tend to have greater awareness of biophilic development.

For individuals intending to adopt biophilic development, Figure 5(a) highlights that the age group 50–59 has the highest intention, followed by those aged 30 and under. In Figure 5(b), individuals with a master’s or Ph.D. degree show the highest intention to adopt biophilic development. Figure 5(c) indicates that senior engineers exhibit the highest intention, but the difference between groups is minimal, with all professional titles showing similar ratios. Similarly, Figure 5(d) shows that individuals with extensive work experience have a slightly higher intention, but the ratios across experience levels are relatively similar. Overall, the differences in IntBD are minimal across engineering titles and work experience, with notable variation primarily observed in age distribution.

The initial results suggest that variations in IntBD are minimal across different professional roles and levels of experience. Instead, age and educational background emerge as the most significant factors driving differences in adoption, with middle-aged individuals and those with advanced degrees showing the highest levels of intention.

Correlation analysis of behavioral factors presents the relationships between three key variables influencing biophilic development adoption: attitudes, subjective norms and PBC. Three scatter plots explore pairwise interactions between these behavioral factors to provide insights into their potential correlations.

The first plot displayed in Figure 6(a): Subjective norms versus PBC reveals a positive relationship between subjective norms and PBC. Data points clustering along a rising trend suggest that stronger subjective norms correlate with an increased sense of control over adopting biophilic development principles. This indicates that social influences, such as peer support or professional expectations, may empower individuals to feel more capable of implementing biophilic practices. In Figure 6(b): Attitudes versus PBC, the scatter plot highlights the connection between attitudes and PBC. Points concentrated in the upper-right quadrant show that individuals with positive attitudes toward biophilic development also tend to feel a higher level of control in adopting such practices. This suggests that a favorable mindset enhances confidence in the ability to implement biophilic principles, reinforcing the interplay between personal beliefs and self-efficacy. Finally, in Figure 6(c): Attitudes versus subjective norms examine the relationship between attitudes and subjective norms. A moderate positive trend is evident, indicating that individuals who perceive strong subjective norms are more likely to hold favorable attitudes toward biophilic development. This underscores the importance of fostering a supportive social and professional environment to cultivate positive attitudes.

The three scatter plots collectively demonstrate the interconnectedness of behavioral factors. Strong subjective norms and positive attitudes are linked to higher perceived control, creating a reinforcing effect that could drive the adoption of biophilic development. These findings highlight the need for targeted interventions focusing on social and attitudinal factors to encourage the integration of biophilic principles in engineering practices.

To build on these insights, the data were further analyzed with the inclusion of knowledge as a key variable. The results are illustrated in pair plots, highlighting the differences between two main variables: (1) awareness of biophilic development and (2) intention to adopt biophilic development. These analyses are presented in the next subsection, offering deeper insights into the behavioral and knowledge-based drivers of biophilic development adoption.

The influence of knowledge on the relationships between key behavioral factors, including attitude, subjective norm and PBC is further explored in the context of biophilic development adoption. By analyzing feature interactions through pair plots, the charts compare two key variables: awareness of biophilic development and intention to adopt biophilic development. These visualizations aim to uncover patterns and differences that provide deeper insights into how knowledge interacts with behavioral factors, influencing awareness and intention. The analysis highlights critical trends that inform strategies for applying biophilic development adoption in engineering practices. The analysis seeks to highlight differences that can guide strategies to enhance awareness and promote adoption. The visualizations employ pair plots, with diagonal kernel density estimate (KDE) plots illustrating individual feature distributions and scatter plots showcasing feature interactions.

The diagonal KDE plots in Figure 7 illustrate the distributions of knowledge scores, attitudes, subjective norms and PBC, segmented by awareness of biophilic development (AweBD). Each plot uses color-coding, with blue representing respondents AweBD and red representing those unaware. These visualizations reveal distinct patterns across the groups, providing insights into how awareness influences each behavioral factor. For knowledge scores, the distribution shows that respondents AweBD (blue) tend to have slightly higher values compared to those unaware (red). This suggests that awareness correlates positively with higher knowledge levels. Similarly, the KDE plot for attitude reveals overlapping distributions, but those who are AweBD exhibit a slight shift toward higher attitude values, indicating a more favorable perception of biophilic development. The pattern continues with subjective norms, where the distribution for the aware group skews toward higher values compared to the unaware group. This finding suggests that individuals AweBD perceive stronger social influence or support for adopting biophilic principles. Finally, for PBC, the separation between the two groups becomes more distinct, with respondents AweBD showing a greater concentration at higher values. This implies that awareness is strongly associated with an increased sense of control over implementing biophilic practices. The diagonal plots highlight that respondents AweBD generally score higher across all behavioral factors, reinforcing the importance of knowledge and awareness in fostering positive attitudes, stronger subjective norms and greater perceived control. These findings provide a foundation for exploring how feature interactions contribute to awareness and intention in subsequent analyses.

The off-diagonal scatter plots provide further insights into the relationships between these variables. The plot for attitudes versus subjective norms shows a positive correlation for both groups, but respondents who are AweBD tend to cluster in the upper-right quadrant, indicating a stronger alignment between these two factors. Similarly, the relationship between PBC and both attitudes and subjective norms demonstrates a positive trend, with blue points predominantly appearing in higher-value regions. This pattern suggests that respondents with awareness not only exhibit stronger attitudes and subjective norms but also perceive greater control over adopting biophilic development. Overall, the chart highlights the interconnectedness of these behavioral factors, showing that awareness of biophilic development is associated with higher scores across all variables. These results emphasize the importance of raising awareness as a critical step toward fostering favorable attitudes, stronger subjective norms and greater perceived control, all of which are essential for promoting biophilic development adoption.

Figure 8 provides a visualization of the relationships between knowledge score, attitudes, subjective norms and PBC, segmented by respondents' IntBD. Data points are color-coded, with blue representing respondents intending to adopt biophilic development (True) and red representing those not intending to adopt (False). The diagonal KDE plots show the distribution of each feature. Respondents intending to adopt biophilic development consistently score higher across all four features compared to those not intending to adopt. The KDE plot for knowledge score reveals a clear distinction, with blue curves skewed toward higher values, indicating that knowledge positively correlates with the intention to adopt. Similarly, the KDE plot for attitudes shows that respondents with higher attitude scores are more likely to express an intention to adopt biophilic development. The distributions for subjective norms and PBC follow the same trend, with respondents intending to adopt biophilic development clustering at higher values for both factors.

The off-diagonal scatter plots provide additional insights into the pairwise relationships between these variables. For attitudes versus subjective norms, a strong positive correlation is evident for both groups, but respondents intending to adopt biophilic development tend to concentrate in the upper-right region, highlighting stronger alignment between these two factors. Similarly, the interaction between PBC and attitudes shows that higher perceived control aligns with a positive attitude, particularly among those intending to adopt biophilic development. The relationship between knowledge scores and the other behavioral factors also suggests that higher knowledge levels are associated with stronger attitudes, subjective norms and PBC.

The relationship between independent variables and the IntBD was analyzed through both univariate and multivariate logistic regression, displayed in Table 1. The results indicate that attitudes and subjective norms are the most significant predictors of engineers’ IntBD in infrastructure projects. In the multivariate analysis, attitudes showed a significant association with intention (aOR = 3.48, 95% CI: 1.14–8.48, p = 0.006). This means that civil engineers with positive attitudes toward biophilic development are approximately 3.5 times more likely to adopt it in their projects. Similarly, subjective norms demonstrated a significant association (aOR = 2.87, 95% CI: 1.16–7.07, p = 0.022), indicating that social influences, such as expectations from supervisors, colleagues and stakeholders, increase the likelihood of adoption by nearly 2.9 times.

In the univariate analysis, PBC was significantly associated with intention (OR = 4.41, 95% CI: 2.41–8.07, p < 0.001). This suggests that engineers who feel confident in their resources, technical capabilities and ability to overcome challenges are about 4.4 times more likely to adopt biophilic development. However, this significance did not persist in the multivariate analysis, indicating that PBC alone is not sufficient to predict adoption intention when other factors like attitudes and subjective norms are considered.

On the other hand, knowledge (Kscore) did not show a significant association in either univariate analysis (OR = 1.12, 95% CI: 0.83–1.53, p = 0.458) or multivariate analysis. However, since univariate results are included for reference, it is displayed in the table, while multivariate results are omitted due to non-significance.

Further analysis was conducted on the same dataset using machine learning techniques, starting with the random forest model, followed by SHAP value analysis to identify the importance of features influencing individuals’ AweBD and their IntBD. To prepare the data for machine learning, categorical variables were transformed using one-hot encoding. The random forest model was trained with a 70-30 train-test split, achieving an accuracy of 0.78 for AweBD and 0.82 for IntBD, demonstrating its effectiveness in predicting outcomes. Hyperparameter tuning for both tests was performed using grid search and randomized search methods, followed by manual fine-tuning to further enhance model performance, as detailed in Table 2. The final model reflects these optimizations, which are elaborated upon in the subsequent analysis.

The classification report demonstrates high precision, recall, and F1-scores for both random forest models: Class 0 (not aware) predicts individuals who are not AweBD, while Class 1 (intention) predicts individuals with the IntBD. The accuracy of the models is 0.75 and 0.77, respectively, as shown in Table 3. However, the performance for Class 1 in the AweBD model and Class 0 in the IntBD model is comparatively lower, indicating the need for further analysis to improve these predictions.

The confusion matrices and evaluation curves in Figure 9 showcase the performance of two random forest models: one predicting AweBD and the other predicting the IntBD. For the AweBD model, shown in Figure 9(a), the results indicate strong performance in predicting individuals who are not AweBD (Class 0), correctly classifying 86 out of 101 individuals, with only 15 misclassified. However, its ability to predict individuals who are aware (Class 1) is weaker, with 17 correctly classified and 19 misclassified. Similarly, the IntBD model, shown in Figure 9(b), displays performance in predicting individuals with the IntBD (Class 1). It correctly classified 78 out of 93 individuals, with only 15 misclassified as having no intention. However, the model encounters challenges in identifying individuals with no intention (Class 0), correctly classifying 27 out of 44 while misclassifying 17. These results align with the performance metrics detailed in the table.

For the evaluation curve of the AweBD model displayed in Figure 9(c), the (ROC) curve has an area under the curve (AUC) of 0.73, indicating moderate discrimination ability between the two classes. The precision-recall curve, with an average precision (AP) of 0.46, highlights room for improvement in handling class imbalance. In comparison, the IntBD model displayed in Figure 9(d) achieves a higher AUC of 0.77 and an AP of 0.85, demonstrating stronger class discrimination and better performance in predicting individuals with the IntBD, further validating the model’s strengths and limitations.

The RF models identified the most influential features impacting prediction outcomes. The top seven important features from both models are displayed in Figure 10, many of which align well with insights from the earlier analysis. For the AweBD model, as shown in Figure 10(a), the most significant feature is D9B, representing respondents who expressed interest in using biophilic development despite it being their first time hearing about the concept. This underscores the critical role of practical exposure in shaping awareness. The second most important feature is AT4, an attitude-related question asking respondents whether adopting biophilic development would enhance their professionalism. This is followed by the overall average subjective norm, average attitude and average PBC scores, all of which significantly contribute to the model’s predictive power. Other influential features include SN6, a question in the subjective norm domain asking whether professionals in their field support and discuss biophilic development, and D2 Work Exp, which highlights respondents’ years of work experience as a significant factor.

For the IntBD model, as displayed in Figure 10(b), the most influential feature is the average subjective norm score, reflecting the perception that biophilic development is becoming a societal trend in Thailand. This finding emphasizes the importance of collective social influences in driving intention. The second key feature is PC2, a question in the PBC domain, assessing respondents’ confidence in adapting biophilic principles to Thailand’s infrastructure projects. Other significant features include D1 Age and D2 Work Exp, indicating that demographic factors such as age and work experience play an essential role in adoption intentions. Additional important features include PC6, a question in the PBC domain about the capability to successfully manage and implement biophilic development, and SN4, a subjective norm-related question regarding whether professionals in their field are starting to discuss biophilic development. These factors highlight the interplay of personal confidence, social influences and professional experience in shaping decisions.

These findings reveal that AweBD is strongly linked to prior experiences and positive attitudes, while the IntBD is primarily driven by subjective norms and perceived control. The overlapping importance of features such as subjective norms, PBC and work experience underscores the need for strategies that address both individual confidence and social pressures. Interventions aimed at promoting societal trends, enhancing personal confidence and leveraging professional networks could effectively increase both awareness and adoption of biophilic development in infrastructure projects.

However, despite these high predicting results, the model’s performance for Class 1’s AweBD model (awareness) and Class 0’s IntBD (no intention) remained suboptimal. To further refine the analysis, SHAP is advised to conduct a detailed feature importance analysis, providing deeper insights into the features. These insights could inform strategies to enhance model performance and interpretability.

SHAP analysis was utilized to further examine the random forest models and understand how individual features influenced their predictions. This technique decomposes the contributions of different features to the model’s predictions for respondents interested or not interested in biophilic development. By analyzing SHAP values, the decision-making process of the model is clarified at an individual level, highlighting the most impactful factors for each prediction. Summary plots of SHAP values offer a comprehensive view of feature importance across the entire dataset, uncovering patterns and dependencies that might not have been apparent in the initial feature importance analysis provided by the random forest models.

The SHAP summary plot, as shown in Figure 11, visually represents the influence of each feature on predicting project cancellations. The vertical axis lists the features ranked by importance, with the most impactful ones positioned at the top. The horizontal axis displays the SHAP values, which reflect the contribution of each feature to the prediction. Positive SHAP values signify an increased likelihood of a prediction, while negative SHAP values indicate a decreased likelihood. Each dot on the plot represents the impact of a feature on an individual prediction, with colors ranging from purple to orange to reflect the feature’s value. Orange dots correspond to higher feature values, whereas purple dots indicate lower values. The SHAP analysis emphasizes features consistent with the random forest model’s findings, such as attitudes, subjective norms and PBC.

Figure 11(a), which presents the AweBD model, highlights several key factors influencing awareness of biophilic development. The most impactful feature is respondents’ willingness to engage with biophilic design even when encountering the concept for the first time, indicating that openness to new ideas strongly contributes to awareness. A belief that biophilic development enhances professional image, along with exposure to supportive discussions among colleagues or within professional networks, also plays a significant role. Furthermore, respondents with higher confidence in their ability to apply biophilic principles and those with more positive attitudes toward such approaches are more likely to be aware of its relevance. While not dominant, work experience and perceived social expectations also contribute moderately, suggesting that both professional background and peer influence help shape overall awareness.

Figure 11(b), which presents the IntBD model, shows that social influence – such as expectations and support from colleagues or institutions – is the most powerful factor in shaping engineers’ intention to adopt biophilic development. Confidence in adapting biophilic concepts to real-world infrastructure projects also has a strong positive impact, reflecting the importance of perceived ability and control. A broader sense of feasibility and perceived support from societal trends further strengthen the likelihood of adoption. Age also appears to be a notable factor, suggesting that generational perspectives may influence openness to biophilic approaches. In addition, engineers who feel that adopting biophilic principles aligns with professional standards and enhances their image are more likely to show a positive intention. Overall, the SHAP value distribution indicates that higher scores across these factors consistently push the prediction toward adoption.

To better understand these relationships, further analysis using SHAP force plots can help explore how these features affect respondents’ awareness and intention.

The SHAP force plot, as shown in Figure 12, visually demonstrates how individual features contribute to specific predictions, highlighting their impact on the model’s output. The plot illustrates the push-and-pull effects of features, with orange segments representing factors that increase the likelihood of the predicted outcome and purple segments showing factors that reduce the likelihood. The base value serves as the starting point, and the final prediction is determined by adding or subtracting the contributions of each feature.

Figure 12(a) illustrates the SHAP force plot of the AweBD model, where the final prediction (SHAP value = 0.81) reflects the combined effects of individual features that either increase or decrease the likelihood of adopting biophilic development. Positive contributors, shown in orange, include factors such as a strong attitude toward professionalism, supportive subjective norms and high overall attitude scores. These indicate that social and attitudinal factors play a central role in driving the prediction upward. Conversely, negative contributors (depicted in purple) include unfamiliarity with biophilic development and lower PBC. These factors pull the prediction downward, suggesting that limited awareness and lower confidence in one’s ability to adopt biophilic design reduce the likelihood of adoption. This pattern is consistent with the trends observed in the summary plot presented earlier.

Figure 12(b) illustrates the SHAP force plot for the IntBD model. The final prediction, with a SHAP value of 0.47, results from the combined influence of individual features that either increase or decrease the likelihood of adopting biophilic development. One of the strongest positive contributors is the average PBC, which reflects the respondent’s confidence in implementing biophilic principles in real-world projects. Other influential positive features include perceptions of societal trends favoring biophilic development, confidence in adapting these principles to infrastructure contexts and supportive subjective norms. These findings highlight the important role of social and professional influences in shaping intentions toward biophilic adoption. In contrast, negative contributors include unfamiliarity with biophilic development and a lack of prior consideration of biophilic design, both of which suggest that limited awareness and experience can significantly hinder adoption. Additionally, demographic factors such as longer work experience and older age appear to reduce the likelihood of adoption, possibly reflecting the challenges that more experienced professionals may face when adjusting to new design paradigms. This force plot highlights the interplay between supportive factors, such as perceived control and social norms and barriers like unfamiliarity and demographic resistance, offering insights into the factors influencing biophilic development adoption.

The results of this study indicate that, among the surveyed Thai civil engineers from major government agencies, attitudes and subjective norms are significantly associated with their intention to adopt biophilic development principles in infrastructure projects. Engineers who held favorable attitudes toward biophilic developments, such as perceiving them as environmentally beneficial and aligned with sustainable infrastructure goals, were more likely to express an intention to integrate such principles. Similarly, the influence of subjective norms suggests that perceived expectations from colleagues, supervisors or institutions may shape engineers’ willingness to consider biophilic practices. Importantly, PBC emerged as the most influential predictor of intention, underscoring the critical role of engineers’ confidence in their ability to implement biophilic strategies effectively. This finding implies that foster adoption requires not only enhancing individual motivation but also creating enabling conditions such as institutional support, practical tools and technical training that strengthen engineers’ perceived capacity to act. Thus, both internal and external dimensions must be addressed to successfully advance biophilic development in infrastructure projects led by civil engineers.

The strong influence of attitudes and subjective norms on Thai civil engineers’ IntBD can be explained through both established behavioral theory and the specific cultural-organizational context of Thailand. Positive attitudes toward sustainability have been shown to significantly predict environmentally responsible behavior (Ajzen, 1991). In line with this, engineers in our study who perceived biophilic development as beneficial were more likely to express willingness to adopt it. Subjective norms also played a critical role, likely reflecting Thailand’s hierarchical and collectivist work culture, particularly within government agencies, which comprise the majority of this study’s participants. In such settings, engineers tend to align with expectations from supervisors and institutional structures. This deference to authority and emphasis on conformity over individual discretion, alongside a strong adherence to legal and professional standards, underscores the influence of systemic pressure in shaping engineering decisions (Luangkaew, 2019). The study by Guo. and Wang (Guo and Wang, 2024) similarly affirms that social and organizational influence strongly shape behavior in high power distance professional settings, supporting the view that subjective norms are consistently influential in civil engineering practice. These findings underscore the need to not only build positive attitudes but also leverage peer and institutional support in policy strategies targeting behavior change.

The findings of this study indicate that PBC is the most influential factor shaping Thai civil engineers’ intention to adopt biophilic development in infrastructure projects. This suggests that engineers’ confidence in their own abilities and access to institutional support outweigh the impact of positive attitudes or knowledge alone. While the study focuses on engineers who play a central role in design and implementation successful adoption of biophilic principles also depends on engagement from other stakeholders. Policymakers in government agencies are especially critical, as they can translate research insights into actionable design regulations, while end users and local communities contribute grassroots expectations that can serve as social drivers of change. To enhance engineers’ confidence in implementation, support mechanisms should include design manuals, technical prototypes and targeted training programs, particularly within the public sector, where most participants in this study are employed. Turner R (Turner, 2005). found that pilot studies are effective tools for reducing uncertainty in large projects by testing feasibility and guiding risk mitigation strategies, thereby facilitating long-term acceptance within engineering organizations. For sustainable transformation, stakeholder responsibilities should be clearly delineated: the Council of Engineers should integrate biophilic design into continuing professional development (CPD) and support specialized certification; the Ministry of Transport should incorporate biophilic elements into design standards and prototype plans; and local governments should enable community participation to tailor biophilic approaches to local needs ensuring both top-down policy alignment and bottom-up public support.

This study represents one of the first in Thailand to integrate the TPB with explainable artificial intelligence (XAI), specifically SHAP, to examine civil engineers’ behavioral intentions toward adopting biophilic development. The findings from statistical analysis, EDA and machine learning models provide comprehensive insights into the key drivers and barriers influencing awareness and adoption among Thai civil engineers. By combining TPB with SHAP, the study not only strengthens theoretical understanding of behavior but also enhances transparency in model interpretation, a critical consideration in engineering contexts where logical justification and evidence-based reasoning are essential. The explainability of SHAP contributes to trust in machine learning outputs and offers actionable insights for policy and design interventions, especially in high-responsibility professions. Moreover, these insights can inform sustainable infrastructure strategies in developing countries where engineering decisions are often shaped by regulatory and organizational structures. However, this study’s generalizability remains limited, as most participants were licensed civil engineers working in Thai government agencies. The perspectives of private-sector professionals or cross-disciplinary stakeholders were not fully captured. Future research should expand to include engineers in the private sector and conduct comparative studies across ASEAN countries to explore cross-cultural variations in innovation adoption. Longitudinal studies are also recommended to assess the practical outcomes of biophilic development implementation over time.

Manop Kaewmoracharoen: Conceptualization, Methodology, Formal analysis, Writing – Original draft, Funding acquisition. Ariya Aruninta: Conceptualization, Supervision, Writing - Review & Editing, Funding acquisition. Pitcha Jongvivatsakul: Writing – Review and Editing, Supervision. Pantira Parinyarux: Data curation, Formal analysis, Writing – Original draft, Project administration.

This study received ethical approval from the Research Ethics Review Committee for Research Involving Human Research Participants, Group I, Chulalongkorn University (COA No. 211/67) on September 27, 2024.