Incubator mediated university industry collaboration drives entrepreneurship in Iranian agricultural universities Authors: Masoumeh Jaberi, Seyed Mohammad Javad Sobhani, Omid Jamshidi Published: 2026 ABSTRACT Entrepreneurial universities are increasingly acknowledged as pivotal catalysts for socio-economic development, particularly in agriculture, by fostering open innovation ecosystems and cross-sectoral partnerships. Anchored in the Triple Helix framework, this study examines how university–industry collaboration (IC) influences academic entrepreneurship (AE), with a focus on the mediating role of incubator-mediated entrepreneurial activities (IEA). Drawing on data from 127 technology-based startups across three Iranian agricultural universities, a hybrid methodological approach integrating Structural Equation Modeling (SEM) and Artificial Neural Networks (ANN) was employed to capture both linear and nonlinear relationships among key variables. Results indicate that IC is positively and significantly associated with AE. Furthermore, IEA not only mediate this relationship but also independently contribute to entrepreneurial outcomes. ANN analysis revealed IC as the most influential predictor (100% normalized importance), followed by IEA (81.7%). The SEM model explained 53.4% of variance in AE, while the ANN model improved this to 67.7%, demonstrating the added value of machine learning in capturing nonlinear dynamics. This research contributes to the open innovation literature by demonstrating how combining SEM-ANN enriches analytical depth and methodological robustness in higher education studies. It also highlights the strategic function of university incubators in bridging the gap between academic knowledge and market-ready innovations, especially in agri-tech contexts. Practical implications suggest that transforming universities into entrepreneurial institutions requires curriculum reforms, investment in innovation infrastructure, and increased private-sector engagement. These findings offer valuable insights for policymakers and academic leaders aiming to align agricultural higher education with innovation-driven economic growth. Limitations include reliance on self-reported data and a cross-sectional design. Future research could employ longitudinal or mixed-methods approaches to capture dynamic interactions. Introduction Entrepreneurship is widely recognized as a critical engine of economic growth and innovation1,2. As leading institutions in knowledge creation, universities possess significant potential to cultivate entrepreneurial mindsets and capabilities3–5. Grounded in the Triple Helix model of university-industry-government interaction, the role of universities has evolved beyond traditional functions of teaching and research. With the emergence of the “third mission,” universities now actively contribute to regional development and socio-economic progress by embracing the concept of the entrepreneurial university6,7. This transformation is especially vital in agriculture, a sector facing urgent challenges related to food security, environmental sustainability, and rural livelihoods8–10. The entrepreneurial university concept has attracted growing scholarly attention as institutions strive to align academic knowledge with market needs11,12. Within this framework, universities foster entrepreneurship through curriculum reform and experiential learning13, while also commercializing research outputs and supporting technology-based startups14. University-industry collaboration and technology incubators represent two interrelated mechanisms central to this mission15,16. Industry partnerships provide access to capital, market intelligence, and real-world validation, while incubators offer structured support including training, mentorship, and infrastructure to translate academic ideas into viable ventures17,18. Despite their potential, agricultural universities in developing countries like Iran often struggle to operationalize this entrepreneurial vision19. Structural barriers including weak industry linkages, under-resourced incubators, and misaligned institutional incentives hinder effective knowledge commercialization9,20. Moreover, while the Triple Helix model emphasizes university-industry-government synergy21,22, the specific mechanisms through which incubators mediate the relationship between industry collaboration and academic entrepreneurship remain underexplored, particularly in agri-tech contexts13,23. Entrepreneurial universities have emerged as pivotal institutions in contemporary innovation ecosystems, fundamentally transforming from traditional knowledge producers to active economic agents8. This transformation is theoretically anchored in the Triple Helix model, which posits that university-industry-government interactions generate synergistic innovation dynamics22. Concurrently, the open innovation paradigm emphasizes permeable institutional boundaries, where universities externalize research through commercialization while internalizing market insights7. These theoretical frameworks provide the conceptual foundation for understanding how universities can drive agri-tech entrepreneurship, a critical concern given global challenges of food security, climate resilience, and sustainable rural development24,25. While entrepreneurial university concepts have gained traction globally, their manifestation varies significantly across contexts. In advanced economies such as the EU and North America, mature innovation infrastructures support technology transfer. In contrast, Global South contexts including sub-Saharan Africa and South Asia face institutional voids and resource constraints that pose distinct challenges26. Iran’s agricultural sector presents a particularly compelling case. Despite substantial research output in agricultural sciences, commercialization rates remain low, mirroring patterns observed in Kenya’s agricultural universities13,19 and India’s agri-tech ecosystems25. This contextual gap between research capacity and market impact underscores the need for context-sensitive models of academic entrepreneurship. Three interconnected limitations constrain both theoretical understanding and practical effectiveness of agri-tech entrepreneurship in developing contexts. First, despite agriculture’s critical role in food security and rural livelihoods, academic entrepreneurship literature remains disproportionately focused on engineering, business, and health sciences. The unique institutional dynamics of agricultural higher education, including seasonal production cycles, bio-safety regulations, and farmer-adoption barriers, remain underexplored, particularly in Global South contexts. Second, while university technology incubators are widely implemented as entrepreneurship catalysts, their conceptualization often reduces them to passive service platforms rather than active institutional mediators. How incubators translate university-industry collaboration into entrepreneurial outcomes, especially in resource-constrained environments, requires deeper empirical investigation. Third, prevailing analytical approaches such as standalone SEM assume linear relationships among innovation drivers, potentially overlooking the nonlinear, synergistic interactions characteristic of real-world innovation ecosystems. This methodological gap is particularly salient in agricultural contexts where multiple institutional, ecological, and market factors interact complexly. While studies exist in engineering and business contexts, agri-tech entrepreneurship in developing economies like Iran remains underexplored, particularly regarding the mediating role of incubators19. To address these gaps, this study is guided by two research questions. RQ1 asks how university-industry collaboration and incubator-mediated entrepreneurial activities jointly shape academic entrepreneurship in Iranian agricultural universities, and what is the nature of their interaction? RQ2 examines the relative predictive importance of these factors in driving entrepreneurial outcomes, and whether these relationships exhibit nonlinear patterns that linear models might miss? These questions enable us to move beyond descriptive accounts of what works to explanatory insights about how and under what conditions university-industry-incubator linkages generate entrepreneurial outcomes in agri-tech contexts. By answering these questions, our work contributes theoretically to the Triple Helix and open innovation literatures, methodologically through SEM-ANN integration, and practically by offering actionable strategies for strengthening agri-tech ecosystems in resource-constrained settings. This research offers three distinct contributions. It presents the first large-scale empirical investigation of university-industry-incubator dynamics within Iran’s agricultural higher education system, providing insights applicable to similar contexts in the Global South. Theoretically, it breaks new ground by reconceptualizing incubators as active translation mechanisms rather than passive conduits within the Triple Helix framework. Methodologically, it pioneers the use of a hybrid SEM-ANN model in agricultural entrepreneurship research, capturing both linear relationships and complex nonlinear interactions between key factors. Theoretical background and hypothesis development Industry collaboration University entrepreneurship encompasses collaborative actions by students and academics that emerge at the intersection of academic research and commercial activity22. This concept refers to the process of establishing and managing knowledge and technology-based startups rooted in university research11,18. These entrepreneurial activities reflect the transformed role of universities as key actors within the innovation ecosystem. Models such as the Triple Helix emphasize the importance of interactions between universities, industry, and government for innovation6,13,23,25. Despite this potential, academic startups often struggle to connect with industry a key barrier to commercialization. In this regard, Sharifzadeh et al.27, acknowledged in their study that the component of IC is one of the key, driving factors with a positive and significant impact on the development of university entrepreneurship. This foundational variable plays a crucial role in shaping the entrepreneurial university ecosystem, and its activation or enhancement can significantly influence other dimensions and ultimately the overall concept of the entrepreneurial university. Other studies have also identified university-industry linkages as a fundamental and critical factor in the development of entrepreneurial universities, particularly in agricultural higher education institutions9,11. Collaborating with industry stakeholders, learning through experience transfer, networking, and direct interaction among actors are core components of university-industry engagement. Based on these insights, the following hypothesis is proposed: Hypothesis 1 IC has a positive and significant impact on the development of AE. Incubators’ entrepreneurial activity According to the Triple Helix model of innovation, university technology incubators are recognized as key elements of the innovation, technology, and entrepreneurship ecosystem15,28,29. They play a pivotal role in the commercialization of knowledge and the support of nascent businesses17,30. These centers provide nurturing support to core ideas and technological units by identifying promising concepts and fulfilling the needs of emerging ventures16,27. However, the effectiveness of entrepreneurial activities in these incubators often depends on external networks and resources31. Resource dependence theory suggests that industry–university collaboration reduces resource constraints by linking incubators to industrial expertise, market insights, and financial capital32,33. For instance, partnerships with agricultural firms can help incubator teams refine prototypes, access supply chains, or comply with regulatory standards. Numerous researchers including Panakaje et al.34, Samer Abdolamir et al.35, and Gaspar Pacheco et al.7, have highlighted that industry engagement enhances the success rate of incubators by increasing experiential transfer and investment. In Iranian agricultural universities, where incubators are nascent and face systemic challenges (e.g., limited budgets, bureaucratic inertia), university-industry linkages may play a central role in legitimizing incubator activities and fostering sustainable business creation36. By involving industry representatives in mentoring or co-designing incubator programs, universities can ensure that entrepreneurial projects align with market demands, thereby strengthening the impact of incubators. Hypothesis 2 IC has a positive and significant effect on the IEA. Academic entrepreneurship AE acts as an enabler of open innovation processes, reflecting the entry of higher education institutions into the realms of markets and capitalism4,12,21. It plays a central role in transferring knowledge and innovation from academia to economic activity. Accordingly, the university’s mission extends beyond mere knowledge production to include the commercial exploitation of research outcomes28,37. In the innovation ecosystem and in line with the third mission of universities, incubators support entrepreneurial cores and startups, thus facilitating the commercialization of ideas. These initiatives accelerate the development of entrepreneurship within universities35. Beyond the direct role of incubators, education also plays a significant and influential role in strengthening AE. This function is realized through increased student awareness, bridging the gap between theory and practice, and cultivating entrepreneurial skills. Education can also act as an accelerator when combined with stakeholder collaboration34. Empirical findings corroborate these roles—for example, Gaspar Pacheco et al.7, found that initiatives such as incubators and support programs positively facilitate and strengthen AE. This benefit is achieved by reinforcing the connection between research activities and university policies with entrepreneurial actions, ultimately benefiting higher education institutions3. These results suggest that university managers should focus on launching spin-off companies and supporting technology transfer processes to enhance these effective tools28. Therefore, it can be inferred that Iranian agricultural university incubators can facilitate stakeholder participation (e.g., students, faculty, policymakers) in AE, startup formation, and economic growth through supportive programs and investment attraction. Hypothesis 3 IEA have a positive and significant impact on the development of AE. Conceptual framework Based on the theoretical foundations and hypotheses developed above, the conceptual framework of the study is structured around three key constructs (see Fig. 1). In this framework IC directly influences both AE (H1) and IEA (H2). Moreover, IEA, in turn, mediate the relationship between IC and AE, while also exerting a direct effect on AE (H3). This conceptual model reflects the integrated dynamics of the Triple Helix framework, emphasizing the interdependence of universities, industry, and incubators in fostering a sustainable entrepreneurial ecosystem within agricultural higher education. Fig. 1The proposed conceptual Framework. The proposed conceptual Framework. Methodology This descriptive study adopts a quantitative approach with applied objectives. The primary aim is to investigate the impact of IC on the development of AE, while considering the mediating role of IEA within Iranian agricultural universities. The statistical population consists of all CEOs or idea holders from 200 technology-based startups and innovation cores located in the technology incubators of three specialized agricultural universities in Iran: Sari Agricultural Sciences and Natural Resources University (N = 81), Gorgan University of Agricultural Sciences and Natural Resources (N = 70), and Ramin Agriculture and Natural Resources University of Khuzestan (N = 49). A stratified random sampling method with proportional allocation was employed, based on the Krejcie and Morgan table, resulting in a required sample size of N = 127. Of these, 41% were affiliated with Sari University’s incubator (n = 52), 34.6% with Gorgan University’s incubator (n = 44), and 24.4% with Khuzestan University’s incubator (n = 31). This geographical distribution reflects the diversity of university incubators included in the study. Furthermore, the sample was distributed across business development stages. Such that, 65 units were at the pre-growth stage (including ideation and prototyping), and 62 units were at the growth stage (business development), indicating a balanced representation across different phases of startup maturity. These cores and units were selected due to their active engagement within agricultural university incubators, ensuring sufficient familiarity with incubator operations and the capacity to provide high-quality data relevant to the research objectives. This study was conducted in accordance with established ethical standards for research involving human participants. Written informed consent was obtained from all participants prior to their involvement; no minors (individuals under 18 years of age) were included in the study. The research protocol, including the consent procedure, was reviewed and approved by the Ethics Committee of the Agricultural Sciences and Natural Resources University of Khuzestan. To ensure confidentiality, all data were anonymized and stored securely. Participants were also informed of their right to withdraw from the study at any stage without any adverse consequences. Data collection was conducted through an online survey using a researcher-developed questionnaire designed via the Avalform platform (https://form.avalform.com) between January and February 2025. The questionnaire consisted of two main sections. Section One included instructions, the study’s objectives, and demographic information about respondents. Section Two comprised 12 items derived from validated constructs in the literature (see Table 1), measured using a five-point Likert scale. To ensure content validity, the questionnaire was reviewed by faculty members specializing in agricultural extension and education from Sari and Khuzestan agricultural universities. Following expert feedback and revisions, the final version was developed and deployed online. Table 1Measurement items and references.Item*StatementReferenceIC Refers to the extent of collaboration between universities and industry stakeholders in knowledge exchange, joint projects, mentorship, and resource sharingGaspar Pacheco et al.7,Panakaje et al.34,IC1There is frequent interaction between university researchers and industry professionals.IC2The university collaborates with industry partners on joint research and development initiatives.IC3Industry experts participate in mentoring or training programs at the university.IC4The university actively engages in technology transfer and commercialization activities with industry.IEA Refers to the level and quality of support provided by university incubators to startups, including training, networking, funding, and business development servicesGaspar Pacheco et al.7,Panakaje et al.34,IEA1The incubator provides access to business development training and workshops.IEA2The incubator facilitates networking opportunities with investors and entrepreneurs.IEA3The incubator offers guidance and support for product prototyping and market validation.IEA4The incubator supports startups in securing funding and investment opportunities.AE Refers to the overall capacity and performance of the university in fostering entrepreneurship through education, innovation, and startup creationGaspar Pacheco et al.7,AE1The university promotes entrepreneurial thinking among students and faculty.AE2The university encourages the establishment of spin-off and startup companies.AE3The university integrates entrepreneurship into its academic and research programs.AE4The university contributes to the regional economy through entrepreneurial initiatives.*Note All items measured on a 5-point Likert scale (1 = strongly disagree, 5 = strongly agree). Table 1Measurement items and references.Item*StatementReferenceIC Refers to the extent of collaboration between universities and industry stakeholders in knowledge exchange, joint projects, mentorship, and resource sharingGaspar Pacheco et al.7,Panakaje et al.34,IC1There is frequent interaction between university researchers and industry professionals.IC2The university collaborates with industry partners on joint research and development initiatives.IC3Industry experts participate in mentoring or training programs at the university.IC4The university actively engages in technology transfer and commercialization activities with industry.IEA Refers to the level and quality of support provided by university incubators to startups, including training, networking, funding, and business development servicesGaspar Pacheco et al.7,Panakaje et al.34,IEA1The incubator provides access to business development training and workshops.IEA2The incubator facilitates networking opportunities with investors and entrepreneurs.IEA3The incubator offers guidance and support for product prototyping and market validation.IEA4The incubator supports startups in securing funding and investment opportunities.AE Refers to the overall capacity and performance of the university in fostering entrepreneurship through education, innovation, and startup creationGaspar Pacheco et al.7,AE1The university promotes entrepreneurial thinking among students and faculty.AE2The university encourages the establishment of spin-off and startup companies.AE3The university integrates entrepreneurship into its academic and research programs.AE4The university contributes to the regional economy through entrepreneurial initiatives.*Note All items measured on a 5-point Likert scale (1 = strongly disagree, 5 = strongly agree). Measurement items and references. Gaspar Pacheco et al.7, Panakaje et al.34, Gaspar Pacheco et al.7, Panakaje et al.34, *Note All items measured on a 5-point Likert scale (1 = strongly disagree, 5 = strongly agree). Common method bias (CMB) assessment To assess potential common method bias (CMB), which refers to spurious correlations between variables arising from shared method variance, timing, or rater subjectivity, Harman’s single-factor test was performed38,39. Exploratory factor analysis of the 12 measurement items revealed that the largest extracted factor accounted for only 31.23% of total variance, which is below the commonly accepted threshold of 50%. This result indicates no dominant single factor influencing the data and confirms that CMB does not significantly affect the observed relationships, thereby supporting the validity of the findings. Hybrid analytical approach: SEM-ANN To comprehensively explore both linear and nonlinear relationships among exogenous and endogenous constructs, this study employed a hybrid methodology combining SEM and ANN. This dual-method approach allows for a more nuanced understanding of the complex dynamics influencing AE development, overcoming the limitations of univariate modeling techniques. Data analysis was performed using two software packages: SPSS (Version 26.0, IBM Corp., Armonk, NY, USA; URL: https://www.ibm.com/products/spss-statistics) for artificial neural network modeling and descriptive statistics, and SmartPLS (Version 3.3.9, SmartPLS GmbH, Bönningstedt, Germany; URL: https://www.smartpls.com) for partial least squares structural equation modeling. The study was structured into two distinct but complementary stages: Stage 1: Partial least squares structural equation modeling (PLS-SEM) Using SmartPLS software version 3, PLS-SEM was employed to investigate the linear relationships among the constructs. This approach is especially well-suited for research based on nascent theoretical models, as noted by Hair et al.40,. The analytical process consisted of two primary phases. The first focused on evaluating the measurement model to ensure construct reliability and validity. This involved calculating Cronbach’s alpha, composite reliability (CR), and average variance extracted (AVE) to assess internal consistency and convergent validity. Discriminant validity was further examined using the heterotrait-monotrait ratio (HTMT). The second phase centered on the structural model, which was used to analyze the interrelationships among the constructs, with particular emphasis on the mediating influence of IEA. Key fit indices such as R² and Q² were calculated to gauge the model’s explanatory and predictive power. To test the significance and direction of the proposed causal relationships, bootstrapping procedures were conducted, following the methodology outlined by Zeng et al.41,. Additionally, blindfolding techniques implemented in SmartPLS were utilized to generate Q² values for the dependent variables. Despite its advantages, PLS-SEM has recognized limitations in modeling nonlinear interactions and uncovering complex patterns within high-dimensional datasets, as highlighted by Richter and Tudoran42. These constraints informed the decision to proceed with a complementary stage of analysis using alternative methods capable of addressing such complexities. Stage 2: Artificial neural network (ANN) analysis To address the limitations of SEM in capturing nonlinear relationships, a Multi-Layer Perceptron (MLP) neural network was employed. As one of the most widely adopted architectures in ANN modeling, MLP is particularly effective at identifying complex and non-compensatory interactions among variables, even when data do not conform to normality assumptions. The MLP framework implemented in this study consisted of three main components: an input layer, one or more hidden layers, and an output layer. The input layer represented the independent variables IC and IEA while the hidden layer(s) processed complex, nonlinear patterns in the data. The output layer was responsible for predicting the dependent variable, AE. Specifically, the model was configured with a single hidden layer containing two neurons. Both the hidden and output layers utilized the sigmoid activation function, a common choice in classification-oriented research contexts26,43,44. The ANN model was trained using the backpropagation algorithm with a learning rate of 0.1 and a maximum of 1,000 epochs, stopping early if the error stabilized (tolerance = 0.0001). Weights were initialized using the Nguyen-Widrow method to ensure rapid convergence. The mean squared error (MSE) was used as the loss function. The architecture a single hidden layer with two neurons was selected based on empirical testing: we evaluated models with 2 to 10 neurons and found that two neurons yielded the lowest RMSE on the validation set while avoiding overfitting, consistent with recommendations by Xu et al.44, and Liébana-Cabanillas et al.45, for similar sample sizes.To enhance generalizability and prevent overfitting, 10-fold cross-validation was applied during the training process. This approach ensures that the model maintains robust predictive performance across different subsets of the data, thereby increasing confidence in its applicability to real-world scenarios. Table 2 provides a concise summary of the key architectural and training specifications for the MLP neural network used in this study. It details the structural configuration, activation functions, optimization settings, and validation approach employed to model the nonlinear relationships between industry collaboration, incubator activities, and academic entrepreneurship. The parameters were selected to ensure robust learning, prevent overfitting, and enhance the generalizability of the predictive model. Table 2ANN Architecture and Training Parameters.Parameter CategorySpecification/ValueRationale/Reference Network Architecture – TypeMLPEffective for capturing nonlinear relationships in small to medium-sized datasets.– Input Layer Neurons2 (IC, IEA)Corresponds to the two exogenous constructs identified in the SEM analysis.– Hidden Layers1Balances model complexity and interpretability; avoids overfitting.– Hidden Layer Neurons2Optimized via empirical testing (2–10 neurons); lowest RMSE on validation set.– Output Layer Neuron1 (AE)Represents the endogenous construct (AE). Activation Functions – Hidden LayerSigmoidIntroduces nonlinearity; suitable for bounded output prediction.– Output LayerSigmoidEnsures predictions are scaled between 0 and 1 for normalized data. Training Configuration – Training AlgorithmBackpropagationStandard method for supervised learning in MLP networks.– Learning Rate0.1Provides stable convergence without oscillation.– Maximum Epochs1,000Allows sufficient iterations for error minimization.– Early Stopping Tolerance0.0001Halts training if error change is negligible, preventing overfitting. Validation & Generalization – Data Split (Training: Test)80% : 20%Common ratio for maintaining sufficient training data while retaining test robustness.– Validation Method10-Fold Cross-ValidationReduces variance in performance estimation and enhances generalizability.– Performance MetricRMSE, R2Provides interpretable measure of prediction accuracy.– NormalizationMin-Max Scaling to [0, 1]Improves convergence speed and model stability. Table 2ANN Architecture and Training Parameters.Parameter CategorySpecification/ValueRationale/Reference Network Architecture – TypeMLPEffective for capturing nonlinear relationships in small to medium-sized datasets.– Input Layer Neurons2 (IC, IEA)Corresponds to the two exogenous constructs identified in the SEM analysis.– Hidden Layers1Balances model complexity and interpretability; avoids overfitting.– Hidden Layer Neurons2Optimized via empirical testing (2–10 neurons); lowest RMSE on validation set.– Output Layer Neuron1 (AE)Represents the endogenous construct (AE). Activation Functions – Hidden LayerSigmoidIntroduces nonlinearity; suitable for bounded output prediction.– Output LayerSigmoidEnsures predictions are scaled between 0 and 1 for normalized data. Training Configuration – Training AlgorithmBackpropagationStandard method for supervised learning in MLP networks.– Learning Rate0.1Provides stable convergence without oscillation.– Maximum Epochs1,000Allows sufficient iterations for error minimization.– Early Stopping Tolerance0.0001Halts training if error change is negligible, preventing overfitting. Validation & Generalization – Data Split (Training: Test)80% : 20%Common ratio for maintaining sufficient training data while retaining test robustness.– Validation Method10-Fold Cross-ValidationReduces variance in performance estimation and enhances generalizability.– Performance MetricRMSE, R2Provides interpretable measure of prediction accuracy.– NormalizationMin-Max Scaling to [0, 1]Improves convergence speed and model stability. ANN Architecture and Training Parameters. This hybrid SEM-ANN approach ensures robustness in model validation and provides deeper insights into both linear and nonlinear pathways influencing AE within agricultural universities. By integrating rigorous sampling, validity testing, and a dual-method analytical framework, this study enhances the accuracy and depth of findings related to AE development. The combination of PLS-SEM and ANN enables comprehensive exploration of both direct and indirect effects, offering valuable contributions to theory and practice in open innovation ecosystems46. Results Descriptive statistics This descriptive study analyzed data collected from 127 idea holders and technology-based entrepreneurs operating within university incubators in Iran’s agricultural higher education system. The descriptive findings, summarized in Table 3, provide valuable insights into the demographic and professional characteristics of respondents. The majority of respondents were male (72.4%), indicating a dominant presence of men in AE within agricultural incubators. Notably, 82.7% of participants held a Master’s or Doctoral degree, underscoring the high level of educational attainment among participants in these innovation ecosystems. This suggests that agricultural university incubators are effectively engaging postgraduate alumni, potentially contributing to enhanced innovation and competitive capacity in technological projects. The average age of respondents was 38.16 years, with the highest frequency falling within the 36–45 age range often considered the peak period for professional activity and creativity. Additionally, a significant proportion (approximately 40%) of respondents were under the age of 35, reflecting the success of university incubators in attracting emerging entrepreneurs. This diverse age composition fosters synergy by combining youthful energy and innovative ideas with the experience and expertise of older cohorts, thereby enhancing the quality of outputs within university incubation environments. Table 3Demographic distribution.VariableCategoryFrequencyPercentageGenderMale9272.4%Female3527.6%Education LevelDiploma or lower21.6%Associate Degree21.6%Bachelor’s Degree1814.2%Master’s Degree4434.6%Doctoral Degree6148.0%Age Group≤ 25129.5%26–353930.7%36–454535.4%≥ 462519.7%No Response64.7%Total127100% Table 3Demographic distribution.VariableCategoryFrequencyPercentageGenderMale9272.4%Female3527.6%Education LevelDiploma or lower21.6%Associate Degree21.6%Bachelor’s Degree1814.2%Master’s Degree4434.6%Doctoral Degree6148.0%Age Group≤ 25129.5%26–353930.7%36–454535.4%≥ 462519.7%No Response64.7%Total127100% Demographic distribution. Measurement model assessment All factor loadings of the measurement items exceeded the recommended threshold of 0.6, indicating strong item-to-construct relationships. The reliability assessment using Composite Reliability (CR = 0.901–0.945), Cronbach’s Alpha (α = 0.851–0.923), and Rho_A (ρ_A = 0.854–0.924) confirmed high internal consistency across all constructs, with values exceeding the acceptable threshold of 0.740. Composite reliability is preferred over Cronbach’s alpha because it accounts for the inter-construct correlations rather than treating each construct in isolation. Convergent validity was assessed using the Average Variance Extracted (AVE = 0.698–0.812), which surpassed the critical threshold of 0.5, confirming that each latent variable explains more than half of its indicators’ variance47. Model fit was evaluated using the Standardized Root Mean Square Residual (SRMR = 0.070), which fell below the acceptable threshold of 0.08, indicating good model fit. Additionally, the Normed Fit Index (NFI = 0.903) met the recommended benchmark of ≥ 0.90, further supporting the overall adequacy of the measurement model. These results collectively indicate that the measurement model exhibits strong convergent validity and reliability, as detailed in Table 4. Table 4Measurement model evaluation results.ConstructIndicatorFactor LoadingCronbach’s αRho_ACRAVEICIC10.6830.8510.8540.9010.698IC20.877————IC30.885————IC40.879————IEAIEA10.8700.8960.8980.9270.762IEA20.886————IEA30.878————IEA40.857————AEAE10.8910.9230.9240.9450.812AE20.933————AE30.887————AE40.892———— Table 4Measurement model evaluation results.ConstructIndicatorFactor LoadingCronbach’s αRho_ACRAVEICIC10.6830.8510.8540.9010.698IC20.877————IC30.885————IC40.879————IEAIEA10.8700.8960.8980.9270.762IEA20.886————IEA30.878————IEA40.857————AEAE10.8910.9230.9240.9450.812AE20.933————AE30.887————AE40.892———— Measurement model evaluation results. To assess discriminant validity, two approaches were applied. According to Fornell-Larcker48 criterion, the square root of the AVE for each construct was compared with its correlation with other constructs. As shown in Table 5, the diagonal values representing the square root of AVE exceeded the off-diagonal correlations, confirming adequate discriminant validity. Following Henseler et al.49, HTMT values were computed as an alternative criterion. All HTMT values were below the recommended threshold of 0.85, indicating statistically distinct constructs. These findings suggest that the conceptual distinctions among the constructs were well-reflected in the measurement instrument, affirming both statistical and conceptual differentiation. Table 5Discriminant validity matrix.ConstructFornell-Larcker Criterion√AVEHTMT CriterionICIEAAEICIEAIC 0.835* –IEA0.551 0.873 0.628AE0.6470.640 0.901 0.7290.685 *Note: Bolded diagonal values represent the square root of AVE; off-diagonal values indicate correlations between constructs. Table 5Discriminant validity matrix.ConstructFornell-Larcker Criterion√AVEHTMT CriterionICIEAAEICIEAIC 0.835* –IEA0.551 0.873 0.628AE0.6470.640 0.901 0.7290.685 *Note: Bolded diagonal values represent the square root of AVE; off-diagonal values indicate correlations between constructs. Discriminant validity matrix. *Note: Bolded diagonal values represent the square root of AVE; off-diagonal values indicate correlations between constructs. Structural Model Evaluation The structural model was assessed using several key indicators, including path coefficients (β), t-values, R² (the coefficient of determination), and Q² (predictive relevance). To determine statistical significance at the 5% error margin (p ≤ 0.05, t > 1.96), bootstrapping was performed with 5,000 resamples. As shown in Table 6, all three hypotheses received empirical support. The analysis revealed a statistically significant positive association between IC and AE (β = 0.422, t = 4.228, p ≤ 0.001), indicating that higher levels of industry collaboration correspond to higher reported levels of academic entrepreneurship among the sampled startups. Furthermore, IC demonstrated a significant influence on IEA (β = 0.551, t = 6.066, p ≤ 0.001). Lastly, the results confirmed a positive and significant effect of IEA on the development of AE (β = 0.407, t = 3.617, p ≤ 0.001). Table 6Hypothesis testing results.HypothesisPath Coefficient (β)Standard Errort-valuep-value*ResultH1β = 0.4220.1004.228p ≤ 0.001SupportedH2β = 0.5510.0916.066p ≤ 0.001SupportedH3β = 0.4070.1133.617p ≤ 0.001Supported*Note: All p-values are based on two-tailed tests. Bootstrapping was performed with 5,000 subsamples. Table 6Hypothesis testing results.HypothesisPath Coefficient (β)Standard Errort-valuep-value*ResultH1β = 0.4220.1004.228p ≤ 0.001SupportedH2β = 0.5510.0916.066p ≤ 0.001SupportedH3β = 0.4070.1133.617p ≤ 0.001Supported*Note: All p-values are based on two-tailed tests. Bootstrapping was performed with 5,000 subsamples. Hypothesis testing results. *Note: All p-values are based on two-tailed tests. Bootstrapping was performed with 5,000 subsamples. According to Chin’s50 criteria, the R² value of 0.534 indicates a strong predictive power of the model in explaining AE development through IC and incubator activities. Additionally, the Q² value of 0.677, calculated via blindfolding procedures in SmartPLS, confirms the model’s high predictive relevance, surpassing the minimum threshold of Q² > 0. Based on the structural model evaluation, the path coefficients and R² values are illustrated in Fig. 2. In a nutshell, the measurement model demonstrated excellent psychometric properties, including strong reliability, convergent validity, and discriminant validity. The structural model confirmed the positive and significant influence of IC on AE, both directly and indirectly through the mediating role of IEA. Fig. 2PLS-SEM path model showing standardized coefficients (β) and R² values. PLS-SEM path model showing standardized coefficients (β) and R² values. ANN analysis In the subsequent phase of this study, an ANN analysis was conducted using SPSS software, based on the significant variables identified in the SEM stage. The ANN model adopted a single hidden layer architecture, consisting of one output neuron (Representing AE) and two input neurons: (Reflecting IEA and IC). The sigmoid activation function was applied to both the hidden and output layers to ensure nonlinear processing and enhance model adaptability. Input and output values were normalized within the range [0, 1] to improve training efficiency and prediction accuracy45. This structure enabled the model to capture complex, nonlinear relationships between university-industry collaboration, incubator activities, and AE development (see Fig. 3). Fig. 3Proposed ANN model architecture. Proposed ANN model architecture. Following best practices in previous studies involving university settings and technology-based startups42–44, the dataset was split as 80% for training and 20% for testing. To prevent overfitting and ensure robustness, 10-fold cross-validation was employed. The Root Mean Square Error (RMSE) was used as the primary performance metric. Results indicated that the average RMSE was 0.092 (SD = 0.0049) for the training set and 0.0917 (SD = 0.0171) for the test set across the 10 cross-validation folds. The low and consistent RMSE values suggest good predictive accuracy, though the slightly higher standard deviation in the test set indicates some variation in prediction performance across different data subsets. These low RMSE values across all folds validate the high precision and generalizability of the ANN model. The results confirm that the proposed ANN model effectively captures underlying patterns in the data, surpassing the limitations of linear modeling approaches (see Table 7). Table 7RMSE values across folds.RunTraining SampleSSERMSETest SampleSSERMSE11000.9630.098270.1410.07221020.9440.096250.1480.07731030.6880.082240.3260.11741060.8560.090210.2270.10451060.8300.088210.2840.11661080.9110.092190.1640.09371020.8920.094250.2530.10181100.8950.090170.1050.07991010.9420.097260.1390.07310980.8860.095290.2100.085Mean — 0.881 0.0922 — 0.200 0.0917 Standard Deviation — — 0.0049 — — 0.0171 Table 7RMSE values across folds.RunTraining SampleSSERMSETest SampleSSERMSE11000.9630.098270.1410.07221020.9440.096250.1480.07731030.6880.082240.3260.11741060.8560.090210.2270.10451060.8300.088210.2840.11661080.9110.092190.1640.09371020.8920.094250.2530.10181100.8950.090170.1050.07991010.9420.097260.1390.07310980.8860.095290.2100.085Mean — 0.881 0.0922 — 0.200 0.0917 Standard Deviation — — 0.0049 — — 0.0171 RMSE values across folds. Comparative insights: SEM vs. ANN As observed in the PLS-SEM results, the coefficient of determination (R²) for IEA was approximately 30%, while for AE, it reached 53.4%. Additionally, blindfolding results showed positive Q² values indicating good predictive relevance with 0.211 for incubator activities and 0.401 for AE. However, the ANN model significantly outperformed the SEM model in terms of predictive power, achieving explained variances of 47.8% for IEA and 67.7% for AE. While PLS-SEM successfully captured linear associations between IC, IEA, and AE, the ANN model revealed additional nonlinear dependencies, particularly in the influence of IC on incubator activities. Notably, the predictive power of the AE model improved significantly when ANN was employed, underscoring the utility of nonparametric techniques in capturing complexity within open innovation ecosystems. These findings are visualized in Fig. 4. Fig. 4R² values for IEA and AE in ANN model. R² values for IEA and AE in ANN model. To assess the relative importance of each input variable in predicting AE, a sensitivity analysis was performed in Table 8. Normalized importance scores were calculated by dividing each variable’s relative contribution by the maximum value and expressing it as a percentage. Results revealed IC had the highest normalized importance at 100%, identifying it as the most influential predictor. In addition, IEA followed with a normalized importance of 81.7%. These findings align with theoretical expectations and empirical evidence suggesting that industry linkages serve as the backbone of entrepreneurial university development, particularly when supported by effective incubation mechanisms. Table 8Comparative model evaluation: PLS-SEM vs. ANN.ConstructPLS-SEMANNR²Q²Normalized Importance (%)R²IC——100—IEA0.3040.21181.70.478AE0.5340.401—0.677 Table 8Comparative model evaluation: PLS-SEM vs. ANN.ConstructPLS-SEMANNR²Q²Normalized Importance (%)R²IC——100—IEA0.3040.21181.70.478AE0.5340.401—0.677 Comparative model evaluation: PLS-SEM vs. ANN. This comparison highlights the superior predictive capability of the ANN model compared to traditional SEM methods, especially in capturing nonlinear interactions and complex dependencies among latent constructs. The integration of ANN into this research design enhanced the analytical depth and predictive accuracy beyond what is achievable through conventional SEM alone. The ANN model successfully captured nonlinear relationships, validated predictive relevance, and provided insights into variable importance, confirming that: University-industry linkages are the most critical driver of AE.IEA play a pivotal mediating role with substantial independent influence. University-industry linkages are the most critical driver of AE. University-industry linkages are the most critical driver of AE. IEA play a pivotal mediating role with substantial independent influence. IEA play a pivotal mediating role with substantial independent influence. Discussion This study advances understanding of how university-industry collaboration and incubator-mediated activities relate to academic entrepreneurship in Iranian agricultural universities. Grounded in the Triple Helix framework, our findings suggest that industry collaboration is positively associated with entrepreneurial outcomes, and that incubator activities partially explain this relationship. These results align with theoretical expectations while offering context-specific insights for agri-tech ecosystems in resource-constrained settings. Our analysis reveals a dual-pathway pattern in which IC is positively associated with AE both directly and indirectly through IEA, suggesting that industry collaboration may enhance entrepreneurial outcomes through multiple channels simultaneously. This pattern extends prior work by Gaspar Pacheco et al.7, and Romero-Sánchez et al.11, who identified industry collaboration as important but did not examine its mediated mechanisms in agricultural contexts. The direct association between industry collaboration and academic entrepreneurship is consistent with literature emphasizing resource access and market validation4,21. However, the notable mediating role of incubator activities offers a new perspective: incubators may function not merely as service providers but as institutional translators that help convert collaborative potential into entrepreneurial action. Unlike studies conducted in developed economies where universities often possess mature technology transfer infrastructures (e.g15.,), Iranian agricultural universities operate within an environment characterized by fragmented policy support, limited intellectual property enforcement, and developing market linkages9,51. In this context, industry partners appear to serve as more than co-funders. They may act as institutional bridges that provide startups with critical non-financial resources, including market validation through pilot testing, navigation of agricultural regulations, access to distribution channels, and real-time feedback on product usability. This contextual dynamic may help explain why industry collaboration shows both direct and indirect associations with academic entrepreneurship in our sample. Without industry anchoring, even well-designed incubation programs risk producing innovations that remain disconnected from market needs. Second, and more theoretically significant, this study reconceptualizes university incubators not as passive service hubs but as active mediating institutions within the Triple Helix framework. Prior research often treats incubators as outcomes of entrepreneurial university policies (e.g28.,) or as standalone inputs that directly relate to startup success29. In contrast, our model suggests that incubators may function as relational translators. They appear to convert the abstract potential of university-industry partnerships into actionable entrepreneurial support. For instance, when industry partners co-design incubator curricula or participate in pitch evaluations, they may embed market logic into the incubation process, thereby increasing the relevance and viability of spin-offs. This mediating role, supported through both path analysis and ANN sensitivity results, positions incubators as dynamic nodes in the innovation network rather than static infrastructure. Third, the integration of PLS-SEM and ANN reveals patterns that linear models alone might not fully capture. While SEM indicates positive associations between industry collaboration, incubator activities, and academic entrepreneurship, ANN suggests that the marginal impact of incubator support may intensify after a certain threshold of industry engagement is reached. In practical terms, incubators operating in universities with weaker industry collaboration may generate more limited entrepreneurial traction, regardless of their internal quality. Conversely, stronger industry collaboration may enhance incubator effectiveness by providing external legitimacy and resources that reduce uncertainty for early-stage ventures. This finding suggests that the assumption that more incubation universally improves outcomes may require qualification, as incubator effectiveness appears contingent upon the strength of external linkages. Instead, it suggests a contingent effectiveness model where incubator success may be co-determined by the strength of external linkages. Such patterns align with recent calls to treat innovation ecosystems as complex adaptive systems30, where interdependencies rather than isolated components may drive performance. Finally, this study contributes methodologically by demonstrating that hybrid SEM-ANN modeling may offer analytical advantages in open innovation research. While SEM-ANN hybrid approaches have been applied in education and marketing research (e.g43,45.,), this study is among the first to deploy this methodology in the domain of academic entrepreneurship within agricultural higher education. The ANN model’s explanatory capacity suggests that entrepreneurial outcomes in agri-tech may be shaped by synergistic, non-additive interactions among drivers. However, it is important to acknowledge certain limitations of the ANN approach. While ANN excels at detecting complex nonlinear patterns, it operates with limited interpretability of specific parameter relationships compared to SEM’s clear path coefficients. Additionally, the model’s performance is sensitive to sample size. Larger datasets would enhance generalizability and potentially reveal more nuanced interaction patterns42. Future studies could address these limitations by incorporating explainable AI techniques or collecting longitudinal data to validate predictions over time. The ANN sensitivity analysis identified industry collaboration as the most influential predictor, followed by incubator activities. This hierarchy differs from findings in European contexts (e.g18.,) where institutional support often outweighs industry linkages, but aligns with Global South studies24,25 where external validation may be scarcer and thus more valuable. Specifically, in Iran’s fragmented policy environment9, industry partners may provide not just funding but crucial regulatory navigation and supply-chain access that public institutions cannot offer. This reliance on private-sector anchoring appears particularly pronounced when compared to Indian agri-tech ecosystems25, where government programs substantially de-risk entrepreneurship. Furthermore, contrary to technology-focused incubators in business and engineering29, agricultural incubators must bridge disciplinary divides between biotech, data science, and traditional farming, a complexity that may amplify their mediating role. In sum, our findings do more than confirm that collaboration matters. They offer explanatory insights into why collaboration may matter in Iran’s agri-tech landscape, how incubators might transform collaboration into action, and under what conditions these mechanisms may become most potent. This advances the Triple Helix and entrepreneurial university literatures from normative advocacy toward context-sensitive, mechanism-based theorizing. Conclusion and implications Theoretical contributions to academic entrepreneurship This study offers three substantive theoretical contributions to academic entrepreneurship literature. First, we extend academic entrepreneurship theory beyond its business and engineering origins by incorporating agro-ecological specificities such as seasonal production cycles, land tenure complexities, bio-safety regulations, and farmer adoption barriers. These factors may shape entrepreneurial pathways differently than in technology sectors, addressing Chirinda et al.‘s (2024) call for sector-sensitive academic entrepreneurship models. Second, we advance beyond the structural focus of traditional Triple Helix models6 by proposing incubators as active translators that may mediate knowledge flows. This process-oriented perspective responds to Correia et al.,‘s3 critique that most ecosystem studies describe structures but not processes. Third, our findings suggest that incubator impact may be associated with industry linkage strength, challenging universal best practice models. This contingency perspective aligns with recent institutional theory applications in entrepreneurship30 and supports context-sensitive theorizing. Practical implications for agri-tech innovation ecosystems The practical implications of our study offer guidance for strengthening agri-tech innovation ecosystems in contexts characterized by resource constraints and fragmented policy support, such as Iran and other Global South agricultural systems. These recommendations are particularly relevant for agricultural universities operating in environments where formal technology transfer infrastructure is nascent and where industry partnerships serve as critical bridges to market validation. Findings may not generalize to contexts with mature innovation ecosystems where institutional support structures differ substantially. For university leaders in similar settings, this means focusing on cultivating a few deep, strategic industry partnerships rather than many superficial ones, as our evidence suggests this approach may yield superior entrepreneurial outcomes. Concurrently, integrating specialized, agriculture-focused entrepreneurship training covering topics like agribusiness marketing, climate-risk financing, and regulatory navigation into the curriculum may help prepare graduates for the sector’s unique challenges. Incubator managers may amplify their impact by establishing structured co-mentoring programs with seasoned agribusiness professionals, providing startups with guidance on real-world hurdles like seasonal production cycles. Furthermore, instituting field validation protocols, where new technologies are trialed on partner farms prior to scaling, may help ensure market relevance and practical viability. Importantly, this model of collaboration may naturally advance several Sustainable Development Goals. It may support Zero Hunger through precision agriculture tools that boost smallholder productivity, promote Industry and Innovation by fostering inclusive partnerships with small and medium enterprises, and contribute to Climate Action via drought-resilient crops and farm-based renewable energy solutions. Policy and institutional recommendations Given the predictive association of university-industry linkages with entrepreneurial outcomes, we recommend that agricultural universities in similar contexts consider establishing dedicated liaison offices focused on co-designing curricula and research agendas with industry partners. Moreover, since most respondents in our sample held postgraduate qualifications, introducing entrepreneurship education at the undergraduate level may help cultivate an early entrepreneurial mindset and build a sustainable pipeline of entrepreneurial graduates. Open innovation outcomes appear particularly vital in the agricultural sector, where university incubators may drive the commercialization of agri-tech innovations ranging from software applications and smart systems to data-driven methodologies. These innovations may simplify technology adoption for farmers and enhance productivity. Furthermore, collaborative efforts between universities and industries may lead to targeted educational and extension programs grounded in real-world agricultural challenges. Therefore, the framework developed in this study may provide a valuable reference for strengthening the value chain from academia to farm-level implementation in comparable settings. To act as catalysts for innovation and development, agricultural universities may benefit from moving beyond traditional modes of research dissemination and actively engaging with industry stakeholders to identify real-world problems and co-create technology-based solutions. This may require a strategic shift toward demand-driven research, wherein academic projects, theses, and dissertations align with industrial needs. Systems such as the Idea-Need Network can serve as effective platforms for bridging academia, industry, and society, thereby fostering technological development, innovation, and knowledge-based employment. Collaborative mechanisms such as joint conferences, workshops, business roundtables, and field visits may be essential for empowering students, faculty, and industry experts through idea exchange and partnership-building. Ensuring the sustainability of these collaborations may necessitate targeted funding allocations and the provision of financial and non-financial incentives for stakeholders engaged in industry-aligned initiatives. Such measures may help orient research toward market responsiveness and encourage the emergence of entrepreneurial talent. University incubators in the agricultural sector may require skilled personnel and support structures tailored to its unique demands. To meet this need, sufficient financial resources and experienced mentors from the agricultural industry should be mobilized. In addition to revising general entrepreneurship curricula, implementing applied, skill-oriented modules such as courses in agribusiness marketing, smart agriculture, capital acquisition, and regulatory frameworks may significantly enhance the capabilities of aspiring agri-entrepreneurs. Unlike conventional tech startups, agri-tech ventures face distinct challenges, including land access, bio-safety regulations, and alignment with seasonal production cycles. Incubators may adapt their offerings accordingly, providing domain-specific mentorship and facilitating field-testing opportunities to improve startup viability. To promote entrepreneurship and nurture a new generation of agricultural innovators in Iran and similar contexts, entrepreneurship concepts may be embedded throughout academic curricula. Additionally, organizing events such as startup festivals and showcasing successful agri-tech ventures may inspire and educate emerging entrepreneurs. Establishing networks among agricultural entrepreneurs, supported by risk capital investors and specialized funds, may be essential to advance innovative ideas to the stage of commercialization. Ultimately, meaningful growth of university entrepreneurship in agriculture may require universities, government agencies, and industry actors to collaborate in institutionalizing entrepreneurial practices and fostering a culture of innovation. Iran’s national innovation policies such as the Resistance Economy framework and the Fifth and Sixth Development Plans explicitly encourage university-industry collaboration. However, bureaucratic barriers, limited intellectual property protection, and fragmented funding mechanisms may impede effective implementation. Our findings suggest that incubators may serve as policy laboratories to operationalize these national goals at the institutional level, provided they receive regulatory flexibility and dedicated budget lines. Policy integration with sustainable development goals Our findings may directly support several UN Sustainable Development Goals in contexts similar to Iran’s agricultural sector. Specifically, SDG 2 (Zero Hunger) may be advanced through the development of precision agriculture tools that optimize input use and increase smallholder productivity. SDG 9 (Industry, Innovation and Infrastructure) may be supported by fostering inclusive innovation that connects academic research with small and medium agribusinesses. SDG 13 (Climate Action) may be advanced via climate-smart technologies such as drought-resistant crops and carbon-footprint tracking systems. In light of this, our policy recommendations for similar contexts are threefold. First, the creation of Green Innovation Funds may help direct government funding toward university-industry partnerships focused on climate resilient agriculture. Second, the establishment of an SDG Aligned Incubator Certification may help recognize incubators advancing multiple sustainability goals. Third, improved Cross Ministerial Coordination may help align agriculture, education, and environment ministry policies in support of agri-tech entrepreneurship. These recommendations are intended for policymakers operating in resource-constrained agricultural innovation systems and may require adaptation to local institutional conditions. Figure 5 illustrates the theoretical and practical flow from the Triple Helix context through university-industry collaboration (IC), where incubators act as active translators (IEA) to foster academic entrepreneurship (AE). The resulting agri-tech innovations generate specific outputs that contribute to multiple Sustainable Development Goals (SDGs), demonstrating the model’s pathway from knowledge creation to sustainable impact. Fig. 5Conceptual model of incubators as active translators in the agri-tech innovation. Conceptual model of incubators as active translators in the agri-tech innovation. Limitations and future research This study has several limitations that warrant consideration. First, data rely on self-reported perceptions from startup founders, which may introduce response bias despite common method bias controls. Second, the sample is confined to three Iranian agricultural universities, limiting generalizability to other national or sectoral contexts. Findings should be interpreted with caution when applied to settings with different institutional arrangements, resource endowments, or policy environments. Third, the cross-sectional design captures associations but not causal dynamics over time. While our directional hypotheses are theoretically grounded in the Triple Helix framework and resource dependence logic, we acknowledge that temporal precedence cannot be established with the current data. Future longitudinal research is needed to examine whether strengthening industry linkages precedes increases in entrepreneurial activity, or whether successful entrepreneurial outcomes attract greater industry engagement. Fourth, our sample comprises 72.4% male respondents, reflecting broader gender disparities in agricultural entrepreneurship within Iran and similar contexts9. While this distribution mirrors population-level patterns in agri-tech startup leadership, it limits our ability to examine whether the industry collaboration-incubator activities-academic entrepreneurship pathways operate differently for female entrepreneurs. Prior research suggests that women entrepreneurs may face distinct barriers in accessing industry networks and incubator resources37, which could moderate the relationships examined here. We caution against generalizing our findings to female agri-entrepreneurs without further investigation, and we encourage future studies to employ gender-stratified sampling or qualitative approaches to explore potential differences in entrepreneurial pathways. Future research should employ mixed methods to triangulate findings with qualitative insights from industry partners and policymakers. Longitudinal studies tracking startup survival and growth would help clarify temporal dynamics. Comparative studies of agri-tech ecosystems across developing countries such as India, Kenya, and Brazil could test the robustness of these pathways across different institutional contexts. Additionally, exploring the role of government policy as a moderating variable in the Triple Helix framework may deepen understanding of how institutional environments shape entrepreneurial outcomes. Focused conclusion This study examined how university-industry collaboration and incubator-mediated entrepreneurial activities relate to academic entrepreneurship in Iranian agricultural universities. Addressing the first research question, results indicate that industry collaboration is positively associated with academic entrepreneurship, and this relationship is partially explained by incubator-mediated activities. A post-hoc moderation test revealed no significant interaction effect, supporting an additive rather than synergistic model. Addressing the second research question, ANN sensitivity analysis identified industry collaboration as the strongest predictor of academic entrepreneurship, followed by incubator activities, with the ANN model explaining more variance than SEM alone. Theoretically, these findings advance entrepreneurial university literature by reconceptualizing incubators as active translators that may convert collaborative potential into entrepreneurial action, rather than passive service platforms. Methodologically, the hybrid SEM-ANN approach demonstrates enhanced predictive capacity for modeling complex innovation ecosystems. Practically, results suggest that agricultural universities in resource-constrained contexts may benefit from prioritizing deep industry partnerships and strengthening incubator capacity to mediate knowledge commercialization. However, findings are context-specific to Iranian agricultural higher education and may not generalize to settings with mature innovation infrastructures. The cross-sectional design limits causal inference, and the male-dominated sample warrants caution in extending results to female agri-entrepreneurs. Future research should employ longitudinal designs, gender-stratified analyses, and comparative studies across developing economies to test the robustness of these pathways. In sum, this research provides evidence that industry collaboration and incubator activities jointly relate to academic entrepreneurship in agri-tech contexts, offering actionable insights for policymakers and university leaders seeking to align higher education with innovation-driven development in the Global South.