Evaluating logistic regression and geographically weighted logistic regression models for predicting orange-fleshed sweet potato adoption intention in Benin Idrissou Ahoudou1, Nicodeme V. Fassinou Hotegni1, Charlotte O. A. Adjé1, Tania L. I. Akponikpè1, Dêêdi E. O. Sogbohossou1,2, Nadia Fanou Fogny3, Françoise Assogba Komlan4, Ismail Moumouni-Moussa5 & Enoch G. Achigan-Dako1 The low adoption rate of biofortified crops, like orange-fleshed sweet potatoes (OFSP), by farmers remains a major food security concern. Accurate forecasting models for OFSP adoption intention are essential for breeding and introduction projects. This study aims to (i) identify key predictors of OFSP adoption intention among farmers in Benin, integrating various factors, and (ii) investigate regional variations in these predictors through different modeling approaches. We used a diverse set of predictors, including social, geographical, and psychological constructs, to model adoption intention in different sweet potato production areas in Benin. Both logistic regression (LR) and geographically weighted logistic regression (GWLR) models were developed and assessed. The GWLR model significantly outperformed the LR model, achieving a validated result of 94.2%, compared to 87% for the LR model. The GWLR model accurately identified areas with medium and high adoption propensities, mainly in northern Benin, aligning closely with observed data. Driving factors showed robust spatial heterogeneities, influencing OFSP adoption intentions differently across regions, with correlations ranging from positive to negative. The GWLR model excels in elucidating the spatial nuances of diverse factors, offering a promising avenue for more reliable predictions for OFSP adoption. Keywords  Adoption, Orange-fleshed sweet potato, Biofortified varieties, Geospatially weighted logistic regression, Psychological constructs, Predictive modeling Deficiency in Vitamin A remains a critical issue for global health, predominantly affecting preschool-aged children and expectant mothers in developing countries. While the World Health Organization estimates that one-third of preschoolers globally are affected by VAD due to insufficient dietary intake, the situation in Benin is particularly severe, with approximately 66% of children under five years suffering from this deficiency1. This rate significantly exceeds the African average of 40%2,3, marking Benin as a critical intervention area. The impact of VAD in Benin shows stark regional disparities. Northern regions report prevalence rates of around 82% of children under five years, with the Alibori and Atacora departments experiencing some of the highest rates nationally1. This geographical variation in VAD prevalence aligns with broader patterns of rural– urban health disparities, where rural areas show consistently higher rates exceeding 70%4. Among pregnant 1Genetics, Biotechnology and Seed Science Unit (GBioS), Laboratory of Crop Production, Physiology and Plant Breeding (PAGEV), Faculty of Agronomic Sciences (FSA), University of Abomey-Calavi (UAC), Tri Postal Cotonou, 01 BP 526, Abomey-Calavi, Republic of Benin. 2International Institute of Tropical Agriculture (IITA), Tri Postal Cotonou, 08 BP 0932, Abomey-Calavi, Republic of Benin. 3Nutrition and Food Systems Unit, Laboratory of Human Nutrition and BioIngredients Valorisation (Nutrifood), Faculty of Agronomic Sciences, University of Abomey-Calavi, Tri Postal Cotonou, 01 BP 526, Abomey-Calavi, Republic of Benin. 4National Institute of Agricultural Research of Benin (INRAB), Tri Postal Cotonou, 01 BP 884, Abomey-Calavi, Republic of Benin. 5Laboratory of Research On Innovation for Agricultural Development (LRIDA), University of Parakou (UP), BP 1269, Parakou, Republic of Benin. email: idrisahoudou@gmail.com; e.adako@gbios-uac.org OPEN Scientific Reports | (2025) 15:8927 1| https://doi.org/10.1038/s41598-025-85173-1 www.nature.com/scientificreports http://www.nature.com/scientificreports http://crossmark.crossref.org/dialog/?doi=10.1038/s41598-025-85173-1&domain=pdf&date_stamp=2025-3-14 women, the situation is equally concerning, with approximately 17.7%5 experiencing VAD and 14.1% suffering from night blindness4. This alarming prevalence highlights the urgent need for effective intervention strategies to address VAD and mitigate its serious consequences. One promising solution to combat vitamin A deficiency (VAD) and other micronutrient deficiencies is biofortification. This innovative strategy enhances the nutritional content of staple crops like rice, wheat, maize, beans, and sweet potatoes6, making them richer in essential nutrients such as iron, zinc, vitamin A, vitamin C, and folate6 through selective breeding6,7, agronomic practices8, and genetic modification6,7. By incorporating these nutrient-dense foods into daily diets, biofortification offers a sustainable, scalable, and cost-effective approach to improving public health and addressing VAD, particularly in vulnerable populations. A notable achievement in biofortification has been the development of orange-fleshed sweet potato (OFSP). Recognized as an effective biofortified alternative, OFSP plays a crucial role in combating micronutrient deficiencies9,10. Consuming 100–125 g of OFSP roots daily can fulfill approximately 100% of the recommended vitamin A intake for children under five years old10. This not only addresses Vitamin A deficiency (VAD) but also aids in preventing sub-clinical forms of the condition. Promoting OFSP and similar biofortified crops is essential in addressing this widespread health challenge. An ex-ante impact assessment indicated that promoting OFSP could impact over 50 million children under five years old in sub-Saharan Africa11. Initiatives to boost the cultivation and consumption of orange-fleshed sweet potatoes (OFSP) continue among rural farming households12,13. These initiatives encompass educating farmers on the nutritional benefits of OFSP and the importance of Vitamin A-rich diets14,15. Additionally, farmers are provided with OFSP vines as planting materials and are taught better management practices16. The goal is to change the perceptions of farm household members about OFSP, ultimately affecting their decisions related to its cultivation and use15. This forms part of a larger strategy to address Vitamin A deficiency, a pressing public health concern in sub-Saharan Africa. Despite the potential benefits of OFSP and the extensive promotional efforts, the adoption of OFSP remains lower than expected17. Studies examining this issue have primarily focused on agronomic and organoleptic factors16,18,19, market dynamics16, seed systems16,17, socio-economic factors that could explain this phenomenon. Some studies, including those by Adekambi et al.14; Jenkins et al.16; Shikuku et al.15, have employed various psychometric models to explore the impact of belief factors on the low adoption rate of OFSP. For instance, Jenkins et al.16 applied Rogers20 innovation stages and found that negative perceptions during implementation phase might cause OFSP rejection during confirmation stage unless the advantages in taste, health, and profitability perceived surpass agronomic challenges. However, there is a lack of studies that evaluate disaggregated constructs items in psychometric models, which could provide a clearer identification of the belief factors influencing the adoption of OFSP. Psychometric models are frameworks for measuring psychological attributes such as beliefs, attitudes, and perceptions that cannot be directly observed21. These models are essential for understanding the abstract factors influencing adoption decisions. Traditional adoption models, such as the Technology Acceptance Model (TAM)22 and Diffusion of Innovations (DOI)20,23, often rely on aggregated measures that combine multiple items into composite scores24,25. While effective for summarizing trends, this approach can obscure the specific effects of individual belief components on adoption decisions21. In contrast to the aggregated approach, disaggregated construct items offer a more granular perspective. This involves breaking down broader psychological elements into specific components26. By treating measurement items separately, this approach provides greater granularity and reveals the distinct impacts of individual factors27,28. For example, Geng et al.29 demonstrated that disaggregated measures enhance predictive accuracy, while Dutta et al.30 highlighted how different dimensions of trust uniquely influence adoption outcomes. In agricultural technology adoption studies, disaggregation has been shown to provide more nuanced insights31, highlighting its ability to refine OFSP adoption strategies. This approach is particularly relevant for identifying the specific drivers of OFSP adoption, such as perceptions of health benefits and agronomic performance, which traditional models may overlook. Socio-environmental and agro-ecological locations are often overlooked in the development of models for new variety adoption, despite their significance highlighted. For instance, Mansaray et al.32 found the positioning of agricultural land (uplands or inland valley swamps) affected how widely improved rice varieties were adopted. Thuo et al.33 highlighted, in their research on new groundnut varieties adoption, that the specific environment where farmers work significantly impacts their acceptance of these varieties. According to the authors, farmer’s location determines opportunities and resources available for technology adoption, playing a pivotal role in Ugandan farmers’ decisions to embrace new technology. This aligns with the ealier observations of Genius et al.34, who suggested that the local context significantly influences farmers’ opportunities, incentives, and responses. Conversely, Conley and Udry35 study showed that farmers’ decisions in adopting new practices are significantly shaped by the insights they gain from their seasoned neighbors, underlining the impact of neighborhood on farmers’ decision-making. In light of these findings, our study aims to bridge critical gaps in understanding OFSP adoption by integrating farmers’ beliefs with geographical insights. While previous studies have acknowledged the role of location in technology adoption, our research represents a methodological advancement by systematically integrating Geographic Information System (GIS) technology with psychological frameworks. Unlike prior research that treated location (region) as a static demographic variable, we employ GIS to capture the dynamic, nuanced ways geographical contexts (local point) shape farmers’ technology adoption decisions. This approach not only enhances the predictive power of existing models but also provides more granular insights into the complex interplay between farmers’ beliefs, local environments, and technology acceptance. Specifically, we expand on existing adoption models by: (1) disaggregating psychological construct items to reveal more precise belief factors; (2) utilizing GIS to capture micro-geographical variations that influence technology adoption; (3) Scientific Reports | (2025) 15:8927 2| https://doi.org/10.1038/s41598-025-85173-1 www.nature.com/scientificreports/ http://www.nature.com/scientificreports integrating psychological frameworks (Technology Acceptance Model (TAM), and Perceived Characteristics of Innovation (PCI)) with spatial analysis. By considering the nuanced psychological elements and the specific socio-environmental contexts, we aim to develop a more comprehensive understanding of OFSP adoption strategies. Our study goes beyond previous research by demonstrating how geographical factors interact with farmers’ perceptions, potentially offering more targeted and context-specific approaches to promoting biofortified crops. Conceptual frameworks and hypothesis This study adopts an integrated approach by combining elements from multiple theoretical frameworks to provide a comprehensive understanding of the factors influencing the adoption of OFSP in Benin. Specifically, we integrate the Technology Acceptance Model (TAM), Innovation Diffusion Theory (IDT), and the Perceived Characteristics of Innovation (PCI) to capture both individual perceptions and broader socio-environmental influences. Technology acceptance model (TAM) The Technology Acceptance Model (TAM) is framework used to understand the factors that influence an individual’s decision to accept and use new technologies36–38. According to TAM, two key factors: Perceived Usefulness (PU) and Perceived Ease-of-Use (PEOU), primarily drive technology adoption. Perceived Usefulness (PU) refers to the degree to which a person believes that using a particular technology will enhance their job performance22,39. In the context of agriculture, PU refers to how farmers perceive the potential of new farming technology or varieties in improving their crop yield or reducing their labor effort. Perceived Ease-of-Use (PEOU) refers to the degree to which a person believes that using the technology will require minimal effort22,40. In agriculture, this could refer to how easy a farmer finds it to implement and use new farming technologies or system. While TAM provides valuable insights into individual perceptions, it does not fully capture the role of the innovation’s characteristics or external factors such as social influences and the broader agricultural context. To address these gaps, we incorporate Innovation Diffusion Theory (IDT). Innovation diffusion theory (IDT) and perceived characteristics of innovation (PCI) Innovation Diffusion Theory (IDT), first conceptualized by Everett Rogers23,41, focuses on how new ideas and technologies spread within a social system42. IDT was formulated to elucidate the reasons individuals decide to embrace or discard an innovation, influenced by their beliefs. The core principle of IDT posits that the process of deciding to adopt an innovation involves five stages: knowledge, persuasion, decision, implementation, and confirmation23,43. The decision-making process results in intention, which is considered a direct predictor of implementation behavior Ajzen44. The author suggests the characteristics of an innovation itself influence the adoption and decision-making process regarding it. These characteristics can be evaluated based on five criteria: relative advantage, compatibility, complexity, trialability, and observability23,43,45. Both TAM and IDT have been extensively applied in various farm-level adoption studies such as conservation production systems46–49, pest management50; precision agriculture and smart farming51–55; new varieties adoption56–61, agriculture extension service62,63, and various ex-ante studies in the field of precision agriculture64–67. Moore and Benbasat68 refined Innovation Diffusion Theory (IDT) by introducing Perceived Characteristics of Innovation (PCI) to shift focus from inherent attributes to adopter perceptions. Modifications included: (i) redefining “advantage” into “relative advantage” and "image," emphasizing social status enhancement69; (ii) dividing “observability” into two components: "result demonstrability," related to tangible outcomes and the ability to communicate benefits, and "visibility," highlighting how easily others can learn from observing innovation use; (iii) introducing the concept of “voluntariness” to acknowledge its impact on an individual’s choice to adopt an innovation69; (iv) redefining the concept of complexity in terms of ease of use. These modifications led to the identification of eight key Perceived Characteristics of Innovation (PCI), namely: i. Relative Advantage: "how much an innovation is perceived as an improvement over its predecessor," ii. Ease of Use: "how simple an innovation is perceived to use," iii. Image: "how much using the innovation is thought to improve one’s social standing," iv. Visibility: "how observable the use of the innovation is within the organization," v. Compatibility: "the degree to which the innovation aligns with the current values, requirements, and past experiences of potential users," vi. Results Demonstrability: "how observable and communicable the outcomes of using the innovation are," vii. Voluntariness: "how much using the innovation is seen as a choice," viii. Trialability: "how much an innovation can be experimented with prior to complete adoption." Integration of technology acceptance model (TAM) and perceived characteristics of innovation (PCI) The constructs utilized in the Technology Acceptance Model (TAM) essentially form a subset of the Perceived Characteristics of Innovation (PCI), as indicated by Moore and Benbasat68 and others. They found a similarity between the relative advantage construct and Perceived Usefulness (PU), and between the complexity (ease of use) construct and Perceived Ease of Use (PEOU). Therefore, this study excludes the constructs of ease of use and relative advantage from the eight Perceived Characteristics of Innovations (PCIs). Moreover, according to Moore and Benbasat68, Voluntariness is a concept suggesting that adopting an innovation is optional. This is especially pertinent in business settings where supervisors might sway the adoption choices of employees. Conversely, in agricultural scenarios, choosing a new variety typically remains a voluntary decision, distinct from corporate influences. As a result, voluntariness is also omitted from the eight primary constructs of interest. Scientific Reports | (2025) 15:8927 3| https://doi.org/10.1038/s41598-025-85173-1 www.nature.com/scientificreports/ http://www.nature.com/scientificreports External factors effect on new technology adoption by farmers In addition to individual perceptions and innovation characteristics, external factors, including socio-economic and socio-demographic characteristics of farmers (age, education, income) and farm attributes (farm size, access to markets, environmental conditions), significantly affect adoption decisions. Studies emphasize the importance of these external influences in shaping adoption behavior. Socio-demographic characteristics Socio-demographic factors like age, education, income, and farming experience play a crucial role in the adoption of new agricultural technologies70–72. For example, Saengavut and Jirasatthumb73 found a positive relationship between socio-demographic variables and technology adoption in Thailand. Similarly, Nyang’au et al.74 identified the importance of income, household size, and access to credit as factors influencing adoption decisions. Pierpaoli et al.75 further emphasized that education level, age, farming experience, and self-confidence shape farmers’ attitudes towards adopting new technologies. These socio-demographic characteristics are often intertwined with psychological factors, such as environmental values, which significantly impact the adoption of sustainable innovations76. For instance, organic farmers, who tend to hold stronger environmental values, are more likely to prioritize sustainability over economic gains, reflecting the importance of these values in driving adoption76. Farm characteristics and geographical context In addition to socio-demographic variables, farm attributes are key determinants of technology adoption. Research has shown that farm size, significantly influence farmers’ likelihood of adopting new technologies71,77,78. In addition, other studies shown that farms located in areas with better access to markets or extension services, or those in regions with specific environmental conditions, might have different propensities for adopting technologies71,79,80. Therefore, geographical location and accessibility are important factors to considered in agricultural technology adoption. These findings underscore the critical role of external factors in shaping agricultural technology adoption dynamics. Rizzo et al.76 highlighted that integrating innovation characteristics with individual psychological and socio-demographic variables can enhance the adoption process. Building on this perspective, incorporating these external influences into the framework is essential for addressing the diverse challenges and contextual nuances farmers face55,76,81. This approach provides a more comprehensive understanding of adoption dynamics, aligning with the theoretical premise that both internal perceptions and external conditions influence adoption decisions. Hypothesis Model Comparison Hypothesis (H1): The Geographically Weighted Logistic Regression (GWLR) model demonstrate better predictive performance for agricultural technology adoption intention compared to the standard Logistic Regression (LR) model. Spatial Heterogeneity Hypothesis (H2): The influence of adoption predictors (psychological, socio-economic, and farm characteristics) on OFSP adoption intention varies significantly across geographical space. Material and methods Study area and sampling method The study was conducted in Benin, a country situated in West Africa, bordered to the northwest by Burkina Faso, to the northeast by Niger, to the west by Togo, and to the east by Nigeria. Benin also has a coastline along the Gulf of Guinea to the south, which is part of the Atlantic Ocean (Fig. 1). The survey was conducted in five departments out of the 12 in Benin, using a purposive sampling approach. These departments include Alibori, Atacora, and Borgou in the northern part of the country, and Atlantique and Ouémé in the southern part. The departments were chosen based on data of sweet potato production over the previous three years, the introduction attempts of orange-fleshed sweet potato (OFSP), and regions identified in the literature as having a high risk of Vitamin A Deficiency (VAD). At the departmental level, municipalities within each department were selected using a simple random sampling method through a lottery-based technique to ensure equal chance for all municipalities. A total of 17 municipalities out of 42 municipalities were surveyed across the five departments. The final stage of the sampling process involved the selection of farmers within the chosen municipalities. Due to the lack of existing databases on sweet potato farmers in each region, compounded by the crop’s limited production among farmers, traditional sampling methods were deemed impractical. Hence, a snowball sampling approach was used, starting with a small group of farmers recommended by the head of each village, who then referred additional participants for inclusion in the study. This process continued until a total of 513 farmers were surveyed across all the municipalities in the five departments. Data collection Data collection occurred between December 2021 and September 2022. Data for this study were gathered via face-to-face interviews using a questionnaire survey with respondents. All participants provided their informed consent before participating in the survey. The survey materials were developed based on an in-depth review of existing literature on the Technology Acceptance Model (TAM) and Innovations Diffusion Theory (IDT) in agricultural sector. The questionnaire used in this study was organized into five sections. The first section collected socio- demographic and socio-economic information from the farmers. The second section aimed to understand the Scientific Reports | (2025) 15:8927 4| https://doi.org/10.1038/s41598-025-85173-1 www.nature.com/scientificreports/ http://www.nature.com/scientificreports significance of sweet potato in production and assess the respondents’ knowledge about orange-fleshed sweet potatoes (OFSP). The third section investigated the social factors that influence the adoption of OFSP. The fourth section included a series of questions designed to evaluate the farmers’ acceptance of the technology (TAM) and perception of the characteristics of the innovation (PCI), featuring 13 items across seven sub-sections: (1) one image item, (2) one compatibility item, (3) three demonstrability of results items, (4) three visibility items, (5) three trialability items, (6) one perceived usefulness item, and (7) two perceived ease of use item. Detailed information on these measurement items is provided in the supplementary material (Table S1). The fifth section recorded the farm, road and market locations using handheld global positioning system (GPS) integrated to the KoboCollect app. A 1–3 Likert scale was principally used for the third and fourth section to rate farmers’ responses ranging from strongly disagree to totally agree. In our context, adoption intention, referring to the farmers’ inclination or willingness to adopt OFSP technology, was the primary dependent variable of interest. It was assessed by examining the direct influence of variables collected across different sections of the questionnaire on this intention. The resulting dataset contained locations for 513 respondents (Fig. 1) farm location across the 17 municipalities surveyed. Data analysis This study involved an integration of various factors that influence decision-making processes regarding the uptake and spread of new technologies. The intricate and diverse nature of these factors required a methodical approach to data processing. The analysis was conducted in the following sequence: 1. Variable Selection: the process began with the identification of the most significant variables that were to be included in the model construction. 2. Base Model Construction: a foundational model was established, with logistic regression serving as the primary model in this study. 3. Geographically Weighted Logistic Regression (GWLR) Model Development: leveraging the base model, we crafted a Geographically Weighted Logistic Regression Model (GWLR), enabling the consideration of spatial variations and localized effects in the analysis. 4. Model Evaluation: a thorough assessment of the models’ performance and quality was conducted to ensure effective capture of the underlying patterns in the data. Fig. 1 .  Distribution of Farm location (starts) during survey in Republic of Benin. Map generated using ArcGIS 10.8 (https://www.esri.com). © I. Ahoudou. Scientific Reports | (2025) 15:8927 5| https://doi.org/10.1038/s41598-025-85173-1 www.nature.com/scientificreports/ https://www.esri.com http://www.nature.com/scientificreports 5. Model Validation: the final step involved a validation process for the models, affirming their robustness and reliability, thereby ensuring the soundness of the findings and conclusions. Variables selection We adopted a multi-pronged approach for variable selection, initiating with an analysis of the correlation matrix to identify highly correlated variables. Such correlations may give rise to multicollinearity issues82,83, with a predefined threshold set at 0.7. Subsequently, we calculated Variance Inflation Factors (VIF) to quantify the extent of multicollinearity among the variables84,85. As per the general guideline, if the Variance Inflation Factor (VIF) values surpass the threshold of five or ten, it suggests that the corresponding regression coefficients may not be accurately estimated due to the presence of multicollinearity84,86. In this study, we omitted variables with a Variance Inflation Factor (VIF) value greater than five. Finally, we employed three different methods: Lasso, Ridge, and Backward Elimination, to select the most relevant variables while penalizing or eliminating less significant ones87,88. These methods are particularly effective at handling high-dimensional data and preventing overfitting. The Lasso and Ridge methods are regularization techniques that can handle high-dimensional data and prevent overfitting89. These methods function by incorporating a penalty term into the loss function throughout the model fitting procedure, which has the effect of shrinking the coefficients of less important variables towards zero90–92. This helps in variable selection by effectively removing variables that do not contribute significantly to the prediction. On the other hand, Backward elimination is one of the most straightforward variable selection techniques. It begins with a model that includes all potential variables. The process involves systematically removing variables one by one from the full model, continuing this removal until only variables that significantly contribute to the outcome remain93–95. To determine which method was most effective for variable selection in our context, we compared the models using several criteria: corrected Akaike Information Criterion (AICc), Bayesian Information Criterion (BIC). This combination of methods ensures that our model includes only the most informative variables, addresses multicollinearity, and enhances the model’s interpretability and predictive performance. Logistic regression model This study uses a binary outcome model to examine farmers’ intention to adopt OFSP. The binary outcomes are defined as follows: if a farmer intends to adopt OFSP, it is coded as (y = 1), and if there is no intention to adopt, it is coded as (y = 0). We denote the probability of a farmer’s intention to adopt OFSP as P . A logistic regression (LR) model is utilized to establish the relationship between the probability of intention to adopt and the predictor variables. The regression model was formulated as follows: ln ( P / (1 − P ) ) = α + βX � (1) Here, P denoted the likelihood of OFSP adopting; X referred to the array of predictor variables; α was the constant term; and β represented the coefficients associated with the predictor variables. After applying a logit transformation, the probability formula for OFSP adoption was derived as follows: P = eα+βX/ 1 + eα+βX � (2) This model allowed for the estimation of the likelihood of OFSP adoption based on the various predictor variables. Geographically weighted logistic regression model The Geographically Weighted Logistic Regression (GWLR) model is an adaptation of the standard logistic regression model that includes spatial location data96. In this model, weighted least square method is used to calculate the parameters for each coordinate point, resulting in local parameter estimation rather than global. As a result, specific parameters correspond to data in different coordinates, this means that data from different coordinates have their own unique parameters. For GWLR model construction, we employed a range of kernel functions, such as Gaussian, Bisquare, and Exponential, to investigate diverse bandwidth options. The model’s optimal bandwidth identified using a selection process that utilized the corrected Akaike Information Criterion (AICc). The assumption made was that, at a specific location (i), the probability of a farmer adopting OFSP (y = 1) is denoted as P, while the probability that they will not adopt OFSP (y = 0) is represented by 1—P. The advantage of using GWLR approach is that it enables us to estimate the likelihood of OFSP adoption based on a variety of predictive variables while also considering their geographical locations. This is particularly useful in studies like ours where geographical factors can have a significant impact on farmers’ decisions. The GWLR model can be expressed as follows: ln ( Pxj / 1 − Pxj ) = α0 (ui, vi) + m∑ k=1 αk (ui, vi) xjk + εi � (3) here, Pxj is the probability of the jth farmers and the function αk (ui, vi) represents the coefficients of the k variables in the model, which exhibit variation according to the region i of latitude and longitude coordinates (ui, vi). The logit transformation of the formula is following: Scientific Reports | (2025) 15:8927 6| https://doi.org/10.1038/s41598-025-85173-1 www.nature.com/scientificreports/ http://www.nature.com/scientificreports Pxj = e α0(ui,vi)+ ∑m k=1 αk(ui,vi)xjk / 1 + e α0(ui,vi)+ ∑m k=1 αk(ui,vi)xjk � (4) The assessment of spatial non-stationarity in the association between adoption intention and predictors was conducted, following the procedure outlined by Chen et al97. This procedure suggests that if a predictor variable’s estimated coefficient’s interquartile range exceeds its standard error in the logistic regression (LR) model, it indicates significant spatial non-stationarity. Models evaluation We employed a range of criteria to evaluate and compare our model fits. Specifically, we considered corrected Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and the negative two Log-likelihood value (-2LogL) as our key metrics96,98. In each of these criteria, the most favorable model was indicated by the lowest value. These evaluation metrics are calculated as follows: AIC = 2k − 2 ln (L) � (5) AICc = AIC + 2k (k + 1)/n − k − 1 � (6) BIC = −2 ln (L) − k ln (n) � (7) In these formulas, L denotes the maximum likelihood estimate of the model, n is the total number of observations, and k represents the number of parameters within the model. The objective is to select the model with the lowest AICc or BIC value as it best balances the trade-off between model fit and complexity. Model validation To assess the model’s performance, we utilized three key metrics: mean absolute error (MAE), root mean squared error (RMSE), and the area under the receiver operating characteristic (ROC) curve. The area under the ROC curve (AUC), is a measure used to demonstrate a system’s proficiency in distinguishing binary data across a variety of threshold settings99,100. Values for the AUC varies from 0.5, which signifies a poor model fit, to 1, indicating an ideal fit101,102. A well-performing model typically demonstrates a low MAE and RMSE alongside a high AUC. The mathematical expressions used to calculate MAE and RMSE are detailed below: MAE = 1/n n∑ i=1 |yi − ŷi| � (8) RMSE = ( 1/n n∑ i=1 (yi − ŷi)2 )1/2 � (9) where, yi​ represents the actual value for the ith observation, ̂yi​ denotes the predicted value for the ith observation, and n is number of observations. Initially, the global Moran’s Index (Moran’s I) was utilized to evaluate spatial autocorrelation within the dataset. Spatial autocorrelation refers to the degree to which similar values are clustered together in space. This helps determine if there is a pattern of similarity or dissimilarity among neighboring observations, and it also evaluates the spatial autocorrelation of residuals. A lower absolute value of Moran’s I suggests less spatial dependency among residuals, indicating that the model effectively accounts for and explains the spatial patterns in the data. Conversely, a larger Moran’s I would imply a stronger spatial autocorrelation in the residuals, suggesting that the model may not fully capture the spatial structure present in the data. The calculation for the global Moran’s I is formulated as follows: I = n/s ∗ [∑n i=1 ∑n j=1 wij (xi − x) (xj − x) /∑n i=1 ((xi − x))2 ] � (10) where: n is the total number of spatial units, for spatial units i and j, xi​ and xj denote the respective residual values, x is the mean of all residuals, s represents the cumulative sum of all spatial weights, and wij ​ is an element within the spatial weight matrix. wij ​ ​ = 1 if spatial units i and j are neighbors, otherwise wij ​ = 0. Moran’s I values can vary from [− 1,1]. A Moran’s I value above 0 implies that the residuals are positively correlated spatially, while a value below 0 suggests a negative spatial correlation. The ROC curve is typically illustrated by plotting the True Positive Rate (TPR) versus the False Positive Rate (FPR) across various threshold settings. TPR, also referred to as sensitivity or recall, and FPR, also called fall-out, are defined in the following manner: T P R = T P/T P + F N � (11) and F P R = F P/F P + T N � (12) Scientific Reports | (2025) 15:8927 7| https://doi.org/10.1038/s41598-025-85173-1 www.nature.com/scientificreports/ http://www.nature.com/scientificreports Here, TP is the count of true positives, FN is false negatives, FP is false positives, and TN is true negatives. Additionally, for a more detailed evaluation of the model’s effectiveness, we calculated precision and the F1-score using the following formulas: P recision = T P/T P + F P � (13) F 1score = 2 ∗ (P recision ∗ Recall)/P recision + Recall � (14) Alongside these metrics, we also measure the accuracy of different models on both training and validation datasets using the formula: Accuracy = T P + T N/n � (15) where, n is the number of observations. In this study, R was used for models fitting and validation. The glm function from the stats package was used for Logistic Regression (LR) model parameters estimation. For the Geographically Weighted Logistic Regression (GWLR) model, bandwidth selection and parameter estimation were carried out using the bw.ggwr and ggwr. basic functions from the GWmodel package. ArcGIS version 10.8 (https://www.esri.com) was used to generate all the maps presented in this study. Results Farmers socio-economic characteristics Most respondents (Table S2) in this survey were men, constituting 87.9% of the sample. The age distribution was concentrated around the middle-age group, with a majority (60.8%) of the farmers aged between 31 and 50 years. A closer look at the farming practices revealed that a substantial majority (69.4%) of farmers operated on smaller fields of 0—5 hectares. Furthermore, 41.5% of farmers allocated approximately 0.26 to 0.5 hectares for sweet potato cultivation. The household size varied, with most farmers (65.5%) having a medium-sized household of between 5 and 9 members. The educational attainment among the respondents was relatively low, with a significant proportion (47.6%) having no formal education. In terms of income, there was a wide range, but it is important to note that a significant proportion of farmers (35.7%) fell into the lowest income bracket (0—60$). This underscores the economic challenges faced by many sweet potato farmers and highlights areas where targeted interventions could be beneficial. Variable selection and logistic regression parameter estimation The Variance Inflation Factor (VIF) values for the various collected variables were less than 5. This suggests an absence of multicollinearity among these factors, as detailed in Table S3. Subsequently, from the remaining variables, the three selection procedures were conducted. The Backward Elimination method outperformed both Lasso and Ridge methods according to all three criteria: it had lower AICc and BIC values (Table S4). All the selected variables in the Backward Elimination method demonstrate statistical significance (P < 0.05) (Table 1). Notably, Farmer Information Reliability (FIR) and Water Source Proximity (WSP) were found to have negative effects on farmers’ adoption intentions. Geographically weighted logistic regression parameter estimation Global Moran’s I analysis of variables selected (Table S5) suggest positive and statistically significant spatial autocorrelation. Assessing the parameter estimates’ spatial stationarity within the GWLR model across the complete dataset (Table S6) demonstrates that all selected variables exhibit non-stationary spatial characteristics excepted for Age. Across the entire study area, the estimated parameters exhibit a spectrum of both positive and negative correlations (Table 2). To capture the nuanced variations in these parameters at local level, we implemented a segmentation strategy, dividing each parameter into five intervals using the Jenks natural breaks classification method. Subsequently, spatial interpolation, illustrated in Fig. 2, was applied to these segmented parameters with ArcGIS version 10.8. The maps displaying t-values for parameter estimates further highlight the locally significant impact of all variables in the GWLR model on the intention of farmers to adopt OFSP (Fig. 3). The coefficient distributions (Fig. 2) unveil distinct spatial patterns in intention. Particularly, the southern part of the country exhibits pronounced spatial heterogeneities, surpassing those observed in other regions. These spatial nuances highlight the varied influence of each assessed variable across different regions (departments). In the extreme southern region, including Atlantique, Ouémé departments, spatial variations for “Production Initiation Ease (PIE)” are relatively modest, while other variables show significant variations. Conversely, in regions situated in the northern part, such as Alibori, Atacora, and Borgou departments, spatial variations for the assessed variables fluctuate between smaller and medium magnitudes. Model evaluation Table 3 displays the fitting and validation statistics for the LR and GWLR models. In the fitting phase, the GWLR model displayed the lowest values for AIC (88.82), BIC (118.6), and -2LogL (58.15), in contrast to the higher values seen with the LR model. Likewise, throughout the validation phase, the GWLR model outperformed in all evaluation metrics. Figure 4A visually represents the ROC curves, aligned with the AUC results, for both models. Figure  4B highlights a distinctive residual pattern observed in the two models. The GWLR model exhibits residuals concentrated around zero, indicating accurate predictions closely aligned with observed values. This concentration suggests the GWLR model effectively captures underlying patterns. In contrast, LR Scientific Reports | (2025) 15:8927 8| https://doi.org/10.1038/s41598-025-85173-1 www.nature.com/scientificreports/ https://www.esri.com http://www.nature.com/scientificreports model residuals are more spread out, signaling higher variability in prediction errors. This spread implies the LR model might be less effective in capturing underlying data patterns compared to the GWLR model. Prediction of OFSP adoption probability in Benin The intention to adopt OFSP was forecasted using the Logistic Regression (LR) and Geographically Weighted Logistic Regression (GWLR) models. Following this, the Kriging interpolation technique was utilized to examine the spatial dispersion of this predicted probability (Fig. 5). Drawing from Staal et al.103, the probabilities P were categorized into three levels for interpretation: low probability (< 0.3), indicating minimal adoption likelihood; medium probability (0.3≤ P  ≤ 0.6), representing moderate adoption potential; and high probability (> 0.6), suggesting strong adoption probability. The LR model predictions reveal a substantial presence of both high and moderate probabilities for OFSP adoption intention across the northern and southern regions of Benin. Notably, in the northern region (Alibori, Atacora, Borgou), these probabilities form a distinct corridor extending from the north-west (Atacora) to the north-east (Alibori, Borgou) (Fig. 5A). In the southern part, while high probabilities were initially observed in Ouémé department, the interpolated results suggest these patterns extend into neighboring Zou and Collines departments. In contrast, the GWLR model underscores that areas with a heightened likelihood of OFSP adoption intention are primarily concentrated in the North-East parts (Alibori, and Borgou) of the country (Fig. 5B). Some areas in the extreme North and North-West of Alibori department display notably high probabilities. Additionally, the interpolated results suggest sporadic regions with moderate probabilities are identified in the central part of Benin (Collines, Donga). Regarding the southern region (Atlantique, Ouémé), the GWLR model uncovers sporadic zones with both moderate and high probabilities, particularly in the extreme southern part of the country (South part of Atlantique department). This pattern suggests a more dispersed distribution of adoption intention in this region. Variables Constructs Min 1st.Qu Median 3rd.Qu Max Intercept -22.54 -22.41 -16.78 -13.6 -11.63 Age 0.04 0.05 0.06 0.06 0.07 Sweet Potato Production Frequency (SPPF) 0.88 1 1.08 2.15 2.18 Sweet Potato Market Access (SPMA) 1.63 1.66 2.36 2.49 2.57 Farmer Information Reliability (FIR) -2 -1.91 -1.62 -1.16 -1.14 Water Source Proximity (WSP) -2.85 -2.54 -1.95 -1.85 -1.69 OFSP Production Compatibility (OPC) COM 1.07 1.17 1.41 1.69 1.72 Explanation of Benefits (EB) ResD 0.97 1.11 1.15 1.82 1.87 Obviousness of Benefits (OB) ResD 1.28 1.4 1.41 1.43 1.52 Peer Production Influence (PPI) VIS 0.35 0.54 0.97 1.83 1.87 Perceived Health Improvement (PHI) PU 0.24 0.46 0.86 1.29 1.29 Production Initiation Ease (PIE) PEOU 1.62 1.64 1.81 1.99 2.05 Table 2.  Geographically Weighted Logistic Regression (GWLR) model parameter estimates. COM: compatibility, ResD: Result Demonstrability, VIS: Visibility, PU: Perceived Usefulness, PEOU: Perceived Ease of Use. Variables Constructs Estimates Std. Error Pr( >|z|) Intercept -18.37 3.14 0.000 Age 0.06 0.03 0.028 Sweet Potato Production Frequency (SPPF) 1.4 0.66 0.035 Sweet Potato Market Access (SPMA) 2.35 0.96 0.014 Farmer Information Reliability (FIR) -1.51 0.59 0.011 Water Source Proximity (WSP) -2.06 0.78 0.008 OFSP Production Compatibility (OPC) COM 1.51 0.59 0.01 Explanation of Benefits (EB) ResD 1.28 0.47 0.006 Obviousness of Benefits (OB) ResD 1.43 0.47 0.002 Peer Production Influence (PPI) VIS 1.16 0.41 0.004 Perceived Health Improvement (PHI) PU 1.08 0.52 0.038 Production Initiation Ease (PIE) PEOU 1.76 0.42 0.000 Table 1.  Logistic regression (LR) model parameter estimates. COM: compatibility, ResD: Result Demonstrability, VIS: Visibility, PU: Perceived Usefulness, PEOU: Perceived Ease of Use. Scientific Reports | (2025) 15:8927 9| https://doi.org/10.1038/s41598-025-85173-1 www.nature.com/scientificreports/ http://www.nature.com/scientificreports Fig. 2.  Coefficients distribution for each variable in the GWLR model. SPPF: Sweet Potato Production Frequency, SPMA: Sweet Potato Market Access, FIR: Farmer Information Reliability, WRP: Water Source Proximity, OPC: OFSP Production Compatibility, EB: Explanation of Benefits, OB: Obviousness of Benefits, PPI: Peer Production Influence, PHI: Perceived Health Improvement, PIE: Production Initiation Ease. Map generated using ArcGIS 10.8 (https://www.esri.com). © I. Ahoudou. Scientific Reports | (2025) 15:8927 10| https://doi.org/10.1038/s41598-025-85173-1 www.nature.com/scientificreports/ https://www.esri.com http://www.nature.com/scientificreports Fig. 3.  t-Values distribution for each variable in the GWLR Model. SPPF: Sweet Potato Production Frequency, SPMA: Sweet Potato Market Access, FIR: Farmer Information Reliability, WRP: Water Source Proximity, OPC: OFSP Production Compatibility, EB: Explanation of Benefits, OB: Obviousness of Benefits, PPI: Peer Production Influence, PHI: Perceived Health Improvement, PIE: Production Initiation Ease. Map generated using ArcGIS 10.8 (https://www.esri.com). © I. Ahoudou. Scientific Reports | (2025) 15:8927 11| https://doi.org/10.1038/s41598-025-85173-1 www.nature.com/scientificreports/ https://www.esri.com http://www.nature.com/scientificreports Discussion In comparison to Linear Regression (LR), Geographically Weighted Logistic Regression (GWLR) demonstrated substantial improvements in model fitting and validation outcomes. GWLR notably excelled in prediction accuracy, as evidenced by consistent results across ROC curves, Moran’s I index, and residual spread. Its efficacy stems from the ability to account for spatial variations, unlike LR, which assumes a uniform relationship between variables across the study area96,104. This assumption neglects the complex interactions between agronomic, socioeconomic, and geographical factors, resulting in a static relationship that overlooks spatial dependencies within adoption zones. The GWLR model addresses this limitation by estimating parameters locally, allowing for region-specific insights. For instance, factors like Peer Production Influence (PPI) and Perceived Health Improvement (PHI) show stronger impacts in southern regions such as Atlantique and Ouémé departments compared to the north. Fig. 4.  The receiving operating characteristic (ROC) curves (A), and residual distributions (B) of LR and GWLR models. Model LR GWLR Fitting metrics AICc 129.81 88.82 BIC 174.2 118.6 -2logL 103.6 58.15 Validation metrics Precision 0.92 0.99 Recall 0.91 0.93 F1-score 0.92 0.96 Moran’s I 0.11 0.08 MAE 0.14 0.1 RMSE 0.29 0.19 AUC 0.87 0.99 Accuracy (%) fit 94.9 95.3 Validation 87 94.2 Table 3.  Performance of Logistic Regression (LR) and Geographically Weighted Logistic Regression (GWLR) Models. AICc: Akaike Information Criterion with correction, BIC: Bayesian Information Criterion, logL: log-likelihood, MAE: Mean Absolute Error, RMSE: Root Mean Squared Error, AUC: Area Under the Receiver Operating Characteristic Curve. Scientific Reports | (2025) 15:8927 12| https://doi.org/10.1038/s41598-025-85173-1 www.nature.com/scientificreports/ http://www.nature.com/scientificreports These location-specific effects enhance not only the model’s predictive accuracy but also its practical relevance, aligning more effectively with real-world adoption patterns. This capability is crucial in agronomy, where factors like soil type, climate, and local practices vary widely105,106. Moreover, this regional specificity ensures better alignment with intervention planning in heterogeneous environments. Household spatial location and farmland resources significantly influence farmers’ decisions to adopt new technologies107,108, similar to the impact of land type on adopting new rice varieties in Central Nepal109. By integrating spatial heterogeneity, the GWLR model not only advances the theoretical framework of adoption models but also provides region-specific and more robust predictions compared to traditional models like LR. This improvement in predictive accuracy is coupled with actionable guidance for tailoring intervention strategies to regional conditions. Additionally, this study addresses concerns raised by Areal and Pede110 regarding unobserved spatial heterogeneity among Filipino rice farmers. They underscore the importance of integrating these effects within frameworks like stochastic frontier analysis for precise efficiency estimates and farm rankings. Their findings emphasize that incorporating spatial variation can guide strategic interventions in agronomy, ensuring tailored solutions for region-specific production challenges. For example, strategic planning based on location can mitigate production challenges in specific regions, as shown by Mamiit et al.111 in Central Luzon, Philippines, concerning water access. Comparatively, our logistic regression (LR) base model predicts orange-fleshed sweet potato (OFSP) adoption probabilities from 0.007 to 0.999. Extending this range slightly from 0.004 to 0.999, the GWLR model provides broader spatial predictions across various regions. In northern areas like Atacora and Alibori, the LR model indicates medium to high adoption probabilities, with higher probabilities localized to parts of the Ouémé department, and segments of Colline, Zou, and Plateau. In contrast, GWLR identifies medium and high adoption probabilities across all northern departments, particularly concentrated in the northeast including Alibori, Borgou, and Donga (Fig. 5). This expanded adoption forecast from the GWLR model is consistent with other studies. Houeninvo et al.112 on improved maize variety adoption in Benin reported a 75% adoption rate in Northern regions, compared to 42% and 29% in the Central and Southern regions, respectively, aligning with our findings on OFSP adoption favoring the north. This regional disparity was attributed to better access to extension services in the North (65%) versus the Central (59%) and Southern (54%) regions, where peer influence on adoption decisions also played a significant role, consistent with observations by Conley and Udry35 in Ghana and Genius et al.113 in Greece. Fig. 5.  Spatial variation of OFSP adoption intention probabilities in Benin, (A) Probability predictions from the LR model, (B) Probability predictions from the GWLR model. Map generated using ArcGIS 10.8 ​(​​​h​t​t​p​s​:​/​/​w​ w​w​.​e​s​r​i​.​c​o​m​​​​​)​. © I. Ahoudou. Scientific Reports | (2025) 15:8927 13| https://doi.org/10.1038/s41598-025-85173-1 www.nature.com/scientificreports/ https://www.esri.com https://www.esri.com http://www.nature.com/scientificreports The adoption of OFSP by farmers is influenced by various factors, with most showing a positive effect. However, "Farmer Information Reliability" (FIR) and "Water Source Proximity" (WSP) stand out as exceptions, negatively impacting adoption. The role of FIR highlights the importance of trust in intra-farm networks as key communication channels in farmers’ decision-making processes, a factor extensively documented across agricultural contexts108,114–116. Farmers often demonstrate higher trust in specific sources, such as agricultural extension agents or researchers compared to peers, even when the latter’s information is accurate. Studies emphasize the substantial influence of trust in extension services on the adoption of new technologies117,118, underscoring the need for institutional involvement. Addressing these barriers by enhancing the reliability of peer information through formalized extension services could help alleviate the adoption challenges posed by FIR. Additionally, waterlogging risks in lowland regions negatively affect OFSP growth119,120. Similarly, in regions like Mozambique, flooding restricts crop production and limits OFSP adoption16. Mitigating these climatic challenges through better drainage systems or the use of flood-resistant crop varieties could further reduce barriers to adoption. This aligns with findings from Ezin et al.121, who noted that waterlogging, especially in areas prone to river overflow like the Ouémé department, significantly constrains sweet potato production. "Production Initiation Ease" (PIE), a component of Perceived Ease of Use (PEOU), emerges as a vital determinant. This underscores the importance of farmers perceiving new technologies as easy to use and integrate into their current operations. The simplicity of integrating new technologies into existing operations positively correlates with adoption122–124. Similarly, OFSP Production Compatibility (OPC) suggests that alignment with existing farming practices significantly enhances adoption likelihood67,125,126. Aligning interventions with existing farming practices and emphasizing compatibility encourages OFSP adoption. The significant impact of Result Demonstrability (ResD) suggests that farmers are more likely to adopt OFSP if they can clearly see and understand its benefits for health, income, or productivity. Demonstration farms and pilot projects showcasing economic returns or health improvements could emphasize these benefits. This pattern observed in the adoption of organic farming among Turkish raisin producers127. In southern regions of Benin, where market access is better, showcasing the economic benefits of OFSP adoption can be especially impactful. In contrast, in northern regions where vitamin A deficiency (VAD) challenges are more pronounced, demonstrating OFSP’s role in health improvement may resonate more effectively with farmers. “Visibility” (VIS) supports the concept of social proof in agriculture, where observing successful OFSP cultivation among peers encourages adoption. This underscores the pivotal role of community and network influences in adoption decisions125,128,129. In southern regions, where market access and community networks are robust, peer success could amplify adoption through strong social influence. This finding aligns with patterns observed in other agricultural contexts. Through “Perceived Usefulness” (PU), specifically "Perceived Health Improvement" (PHI), farmers are more likely to adopt OFSP if they believe it improves family health. Highlighting that OFSP’s nutritional benefits emerge as a key motivator for adoption, as studies show these perceived health enhancements drive adoption16,130. Regarding farmer demographics, age significantly influences farmers’ willingness to adopt OFSP, consistent across various studies in Ghana18, Uganda131, and Mozambique19. This suggests older farmers are more receptive to adopting new varieties like OFSP due to their appreciation of long-term benefits such as nutritional value and food security. The positive impact of "Sweet Potato Production Frequency" (SPF) on adoption intentions is linked to favorable precipitation conditions. Farmers who engage in multiple planting cycles annually have more opportunities to experiment with new varieties like OFSP. This flexibility allows for experimentation, learning, and adjustments across planting cycles. In southern regions, higher annual precipitation (900–1300 mm/year over two seasons) compared to the north (1200 mm/year in one season) creates more opportunities for multiple planting cycles, boosting production frequency and then probably the adoption intentions132. Sweet Potato Market Accessibility” (SPMA) acts as a safety net by enabling farmers to trial OFSP with reduced financial risk. Rodriguez Izaba et al.133 demonstrate that market access significantly enhances the adoption of innovative agricultural technologies. Local markets not only provide direct consumer feedback but also create opportunities for popularizing new varieties134. Strengthening the market value chain and leveraging strong community networks in southern regions could further boost adoption rates. These measures increase can visibility and consumer familiarity, ultimately driving demand over time. The existing strong market for white- fleshed sweet potatoes in the south121,135 could significantly influences OFSP adoption, while in the north, where the focus is on household consumption and supporting the cotton harvest workforce135, adoption dynamics could differ. The influential factors mentioned earlier play a pivotal role in shaping farmers’ inclinations toward adopting OFSP. However, the dynamic interplay of these variables in the decision-making process exhibits a remarkable variance across different regions, adding an intricate layer to the complexity of agricultural choices. The pronounced influence of observed variables in the southern region for nearly all evaluated factors can be attributed, among other factors, to the relatively short distances between fields in the south compared to those in the north. This observation aligns with Tobler136 law of geography "everything is related to everything else, but near things are more related than distant things" is reflected in the southern region, where shorter distances between fields enhance interactions and information sharing. These spatial dependencies not only capture unobserved heterogeneity, such as networks and environmental factors, but also directly influence adoption decisions by creating localized feedback loops that facilitate innovation diffusion. Areal and Pede110 demonstrated in their study on rice production systems that when fields are sufficiently close, small social networks naturally emerge, facilitating information sharing and influencing individual decisions. These spatial dependencies capture unobserved spatial heterogeneity, encompassing the effects of networks and environmental or climatic factors. Similarly, Ambali et al.59 and Houeninvo et al.112 emphasized Scientific Reports | (2025) 15:8927 14| https://doi.org/10.1038/s41598-025-85173-1 www.nature.com/scientificreports/ http://www.nature.com/scientificreports that failing to account for such spatial variations can lead to inefficient or biased analyses, underscoring the importance of integrating spatial heterogeneity into intervention designs. Conclusion In this study, to evaluate the intention to adopt orange-fleshed sweet potato in Benin, we developed both a Logistic Regression (LR) model and a Geographically Weighted Logistic Regression (GWLR) model. Our findings reveal significant spatial heterogeneities in the key drivers influencing OFSP adoption. These drivers include production-related factors (Sweet Potato Production Frequency, Water Source Proximity, Production Initiation Ease), market aspects (Sweet Potato Market Access), information and perception factors (Farmer Information Reliability, Explanation of Benefits, Obviousness of Benefits), and social influences (Peer Production Influence, Perceived Health Improvement, OFSP Production Compatibility). Notably, the GWLR model demonstrated superior predictive capabilities compared to the LR model, highlighting its effectiveness in enhancing the accuracy of predicting OFSP adoption intention. This methodological advancement has two key implications: first, it underscores the importance of considering spatial heterogeneity in adoption studies, suggesting that future research should incorporate geographical variations in adoption modeling; second, it reveals that location-specific factors significantly influence adoption patterns, pointing to the need for spatially targeted intervention strategies. Practical implications of our findings suggest that OFSP promotion programs should be tailored to local contexts, considering the varying importance of adoption drivers across regions. For instance, areas with strong peer influence might benefit from demonstration plots and farmer-to-farmer learning approaches, while regions where market access is crucial might require interventions focusing on value chain development. Additionally, breeding programs should consider location-specific preferences and constraints identified through this spatial analysis to enhance adoption success. These insights are essential for guiding new variety breeding and introduction programs, emphasizing the importance of considering location effects for reliable adoption predictions. The findings not only demonstrate the practical utility of geospatial methodologies in optimizing OFSP adoption initiatives but also provide a valuable framework for future prediction modeling endeavors in agricultural innovation adoption studies. Study limitation While the study employed a comprehensive approach, incorporating socio-economic, socio-demographic, Technology Acceptance Model (TAM), and Perceived Characteristics of Innovation (PCI) variables to predict new variety adoption, several limitations warrant consideration. The use of a three-point Likert scale, specifically for TAM and PCI variables, was designed to facilitate responses from farmers. This decision was further influenced by the lower literacy levels encountered among respondents, which posed challenges in explaining the nuances of each level on a five-point scale. However, this choice introduces potential drawbacks. The limited sensitivity of the scale may hinder the capture of nuanced variations in respondents’ attitudes, potentially compromising the precision and discrimination of the measurements. Additionally, the variable selection procedure used has inherent limitations, including sensitivity to initial conditions and the potential exclusion of relevant non- linear relationships. The study’s moderate sample size of 513 observations may impact the generalizability of the findings. Data availability All data generated or analyzed during this study are included in this published article and its Supplementary Information files. Received: 20 June 2024; Accepted: 1 January 2025 References 1. Food and Agriculture Organization of the United Nations. Profil nutritionnel de pays: République du Bénin. (2011). 2. Black, R. E. et al. Maternal and child undernutrition: global and regional exposures and health consequences. Lancet 371, 243– 260 (2008). 3. UNICEF. Vitamin A Deficiency in Children. UNICEF Data: Monitoring the situation of children and women (2023). 4. FAO. Profil nutritionnel du Bénin. 64 (2011). 5. Alaofè, H., Burney, J., Naylor, R. & Taren, D. Prevalence of anaemia, deficiencies of iron and vitamin A and their determinants in rural women and young children: a cross-sectional study in Kalalé district of northern Benin. Public Health Nutr. 20, 1203–1213 (2017). 6. Garg, M. et al. Biofortified crops generated by breeding, agronomy, and transgenic approaches are improving lives of millions of people around the world. Front. Nutr. https://doi.org/10.3389/fnut.2018.00012 (2018). 7. Van Der Straeten, D. et al. Multiplying the efficiency and impact of biofortification through metabolic engineering. Nat. Commun. 11, 5203 (2020). 8. de Valença, A. W., Bake, A., Brouwer, I. D. & Giller, K. E. Agronomic biofortification of crops to fight hidden hunger in sub- Saharan Africa. Glob. Food Sec. 12, 8–14 (2017). 9. Hotz, C. et al. Introduction of β-carotene–rich orange sweet potato in rural Uganda resulted in increased vitamin A intakes among children and women and improved vitamin A status among children3. J. Nutr. 142, 1871–1880 (2012). 10. Low, J. et al. Sweetpotato in Sub-Saharan Africa. In The Sweetpotato (eds Loebenstein, G. & Thottappilly, G.) 359–390 (Springer Netherlands, 2009). https://doi.org/10.1007/978-1-4020-9475-0_16. 11. Kapinga, R. et al. Vitamin-A partnership for Africa: a food based approach to combat vitamin A deficiency in sub-Saharan Africa through increased utilization of orange-fleshed sweetpotato. Chron. Horticult. 45, 12–14 (2005). 12. Girard, A. W., Brouwer, A., Faerber, E., Grant, F. K. & Low, J. W. Orange-fleshed sweetpotato: Strategies and lessons learned for achieving food security and health at scale in Sub-Saharan Africa. Open Agric. 6, 511–536 (2021). Scientific Reports | (2025) 15:8927 15| https://doi.org/10.1038/s41598-025-85173-1 www.nature.com/scientificreports/ https://doi.org/10.3389/fnut.2018.00012 https://doi.org/10.1007/978-1-4020-9475-0_16 http://www.nature.com/scientificreports 13. Low, J. W., Low, J. W., Low, J. W., Low, J. & Thiele, G. Understanding innovation: The development and scaling of orange-fleshed sweetpotato in major African food systems. Agric. Syst. https://doi.org/10.1016/j.agsy.2019.102770 (2020). 14. Adekambi, S. A., Okello, J. J., Abidin, P. E. & Carey, E. Effect of exposure to biofortified crops on smallholder farm household adoption decisions: The case of orange-fleshed sweetpotato in Ghana and Nigeria. Sci. African 8, e00362 (2020). 15. Shikuku, K. M., Okello, J. J., Sindi, K., Low, J. W. & Mcewan, M. Effect of Farmers’ Multidimensional Beliefs on Adoption of Biofortified Crops: Evidence from Sweetpotato Farmers in Tanzania. J. Dev. Stud. 55, 227–242 (2019). 16. Jenkins, M., Shanks, C. B., Brouwer, R. & Houghtaling, B. Factors affecting farmers’ willingness and ability to adopt and retain vitamin A-rich varieties of orange-fleshed sweet potato in Mozambique. Food Secur. 10, 1501–1519 (2018). 17. Kaphaika, C. S., Katengeza, S. P., Pangapanga-Phiri, I. & Chambukira, M. H. More interventions, low adoption: To what extent are the existing seed sources to blame? The case of orange fleshed sweet potato in Central and Northern Malawi. Sustainability 15, 14390 (2023). 18. Adekambi, S. A. et al. Effect of varietal attributes on the adoption of an orange-fleshed sweetpotato variety in Upper East and Northern Ghana. Outlook Agric. 49, 311–320 (2020). 19. Jogo, W., Bocher, T. & Grant, F. Factors influencing farmers’ dis-adoption and retention decisions for biofortified crops: the case of orange-fleshed sweetpotato in Mozambique. Agrekon 60, 445–459 (2021). 20. Rogers, E. M. How does new innovation spread out ? Diffus. Innov. , 5 th ed . 189–191 (2003). 21. Mislevy, R. J. & Bolsinova, M. Concepts and Models from Psychometrics. In Computational Psychometrics: New Methodologies for a New Generation of Digital Learning and Assessment (eds von Davier, A. A. et al.) 81–107 (Springer International Publishing, 2021). https://doi.org/10.1007/978-3-030-74394-9_6. 22. Davis, F. D. perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Q. 13, 319 (1989). 23. Rogers, E. M. Diffusion of Innovations (Free Press, 1962). 24. Kim, G., Shin, B. & Lee, H. G. A study of factors that affect user intentions toward email service switching. Inf. Manag. 43, 884–893 (2006). 25. Ambong, R. M. A. Technology Acceptance Model (TAM): A Framework for Predicting Farmers’ Adoption of Modern Rice Production and Postharvest Technologies. in Cutting-edge Research in Agricultural Sciences Vol. 12 94–104 (Book Publisher International (a part of SCIENCEDOMAIN International), 2021). https://doi.org/10.9734/bpi/cras/v12/2291E 26. MacKenzie, Podsakoff & Podsakoff. Construct measurement and validation procedures in MIS and behavioral research: Integrating new and existing techniques. MIS Q. 35, 293 (2011). 27. Morin, A. J. S. et al. Complementary variable- and person-centered approaches to the dimensionality of psychometric constructs: application to psychological wellbeing at work. J. Bus. Psychol. 32, 395–419 (2017). 28. Edwards, J. R. Multidimensional constructs in organizational behavior research: An integrative analytical framework. Organ. Res. Methods 4, 144–192 (2001). 29. Geng, W., Liu, L., Zhao, J., Kang, X. & Wang, W. Digital technologies adoption and economic benefits in agriculture: A mixed- methods approach. Sustainability 16, 4431 (2024). 30. Dutta, A., Kovid, R. K., Gupta, M. & Sijaria, R. Smart Homes Technologies Adoption: Effect of Social Trust and Risk Perception. In Economic Recovery, Consolidation, and Sustainable Growth (eds Bexheti, A. et al.) 727–739 (Springer Nature Switzerland, 2023). 31. He, Z., Jia, Y. & Ji, Y. Analysis of influencing factors and mechanism of farmers’ green production behaviors in China. Int. J. Environ. Res. Public Health 20, 961 (2023). 32. Mansaray, B., Jin, S. & Horlu, G. S. A. Do land ownership and agro-ecological location of farmland influence adoption of improved rice varieties? Evidence from Sierra Leone. Agriculture 9, 256 (2019). 33. Thuo, M. et al. Effects of social network factors on information acquisition and adoption of improved groundnut varieties: The case of Uganda and Kenya. Agric. Human Values 31, 339–353 (2014). 34. Genius, M., Pantzios, C. J. & Tzouvelekas, V. Information acquisition and adoption of organic farming practices. J. Agric. Resour. Econ. 31, 93–113 (2006). 35. Conley, T. G. & Udry, C. R. Learning about a new technology: Pineapple in Ghana. Am. Econ. Rev. 100, 35–69 (2010). 36. Ursavaş, Ö. F. Technology Acceptance Model: History, Theory, and Application. in Conducting Technology Acceptance Research in Education : Theory, Models, Implementation, and Analysis (ed. Ursavaş, Ö. F.) 57–91 (Springer International Publishing, 2022). https://doi.org/10.1007/978-3-031-10846-4_4. 37. Marangunić, N. & Granić, A. Technology acceptance model: a literature review from 1986 to 2013. Univers. Access Inf. Soc. 14, 81–95 (2015). 38. Karahanna, E. & Straub, D. W. The psychological origins of perceived usefulness and ease-of-use. Inf. Manag. 35, 237–250 (1999). 39. Liesa-Orús, M., Latorre-Cosculluela, C., Sierra-Sánchez, V. & Vázquez-Toledo, S. Links between ease of use, perceived usefulness and attitudes towards technology in older people in university: A structural equation modelling approach. Educ. Inf. Technol. 28, 2419–2436 (2023). 40. Jing, P., Xu, G., Chen, Y., Shi, Y. & Zhan, F. The determinants behind the acceptance of autonomous vehicles: A systematic review. Sustainability 12, 1719 (2020). 41. Montes de Oca Munguia, O., Pannell, D. J. & Llewellyn, R. Understanding the adoption of innovations in agriculture: A review of selected conceptual models. Agronomy 11, 139 (2021). 42. Monjardino, M., Kuehne, G. & Cummins, J. Value-Ag: An integrated model for rapid ex-ante impact evaluation of agricultural innovations in smallholder systems. Exp. Agric. 56, 633–649 (2020). 43. Rogers, E. M. Diffusion of Innovations (Free Press, 1995). 44. Ajzen, I. From Intentions to Actions: A Theory of Planned Behavior. in Action Control: From Cognition to Behavior (eds. Kuhl, J. & Beckmann, J.) 11–39 (SSSP Springer Series in Social Psychology. Springer Berlin Heidelberg, 1985). ​h​t​t​p​s​:​/​/​d​o​i​.​o​r​g​/​1​0​.​1​0​0​7​/​9​7​ 8​-​3​-​6​4​2​-​6​9​7​4​6​-​3​_​2​​​​​​​ 45. Yuen, K. F., Wang, X., Ng, L. T. W. & Wong, Y. D. An investigation of customers’ intention to use self-collection services for last- mile delivery. Transp. Policy 66, 1–8 (2018). 46. Tucker, M., Napier, T. L. & Nyairo, N. M. Assessment of the Utility of the Diffusion Model for Facilitating Adoption of Soil and Water Conservation Production Systems in North America. in Global Degradation of Soil and Water Resources 93–105 (Springer Nature Singapore, 2022). https://doi.org/10.1007/978-981-16-7916-2_8 47. Wens, M. L. K., Mwangi, M. N., van Loon, A. F. & Aerts, J. C. J. H. Complexities of drought adaptive behaviour: Linking theory to data on smallholder farmer adaptation decisions. Int. J. Disaster Risk Reduct. 63, 102435 (2021). 48. Zeweld, W., Van Huylenbroeck, G., Tesfay, G. & Speelman, S. Smallholder farmers’ behavioural intentions towards sustainable agricultural practices. J. Environ. Manage. 187, 71–81 (2017). 49. Lavoie, A. L., Dentzman, K. & Wardropper, C. B. Using diffusion of innovations theory to understand agricultural producer perspectives on cover cropping in the inland Pacific Northwest, USA. Renew. Agric. Food Syst. ​h​t​t​p​s​:​/​/​d​o​i​.​o​r​g​/​1​0​.​1​0​1​7​/​S​1​7​4​2​1​7​0​ 5​2​0​0​0​0​4​2​3​​​​ (2021). 50. Rezaei, R., Safa, L. & Ganjkhanloo, M. M. Understanding farmers’ ecological conservation behavior regarding the use of integrated pest management- an application of the technology acceptance model. Glob. Ecol. Conserv. 22, e00941 (2020). 51. Carli, G., Xhakollari, V. & Tagliaventi, M. R. How to Model the Adoption and Perception of Precision Agriculture Technologies. in Precision Agriculture: Technology and Economic Perspectives. Progress in Precision Agriculture. (eds. Pedersen, S. M. & Lind, K. M.) 223–249 (Springer International Publishing, 2017). https://doi.org/10.1007/978-3-319-68715-5_11. Scientific Reports | (2025) 15:8927 16| https://doi.org/10.1038/s41598-025-85173-1 www.nature.com/scientificreports/ https://doi.org/10.1016/j.agsy.2019.102770 https://doi.org/10.1007/978-3-030-74394-9_6 https://doi.org/10.9734/bpi/cras/v12/2291E https://doi.org/10.1007/978-3-031-10846-4_4 https://doi.org/10.1007/978-3-642-69746-3_2 https://doi.org/10.1007/978-3-642-69746-3_2 https://doi.org/10.1007/978-981-16-7916-2_8 https://doi.org/10.1017/S1742170520000423 https://doi.org/10.1017/S1742170520000423 https://doi.org/10.1007/978-3-319-68715-5_11 http://www.nature.com/scientificreports 52. Dixit, K., Aashish, K. & Kumar Dwivedi, A. Antecedents of smart farming adoption to mitigate the digital divide – extended innovation diffusion model. Technol. Soc. 75, 102348 (2023). 53. Caffaro, F., Micheletti Cremasco, M., Roccato, M. & Cavallo, E. Drivers of farmers’ intention to adopt technological innovations in Italy: The role of information sources, perceived usefulness, and perceived ease of use. J. Rural Stud. 76, 264–271 (2020). 54. Ahikiriza, E. et al. Farmer knowledge and the intention to use smartphone-based information management technologies in Uganda. Comput. Electron. Agric. 202, 107413 (2022). 55. Masi, M., De Rosa, M., Vecchio, Y., Bartoli, L. & Adinolfi, F. The long way to innovation adoption: insights from precision agriculture. Agric. Food Econ. 10, 27 (2022). 56. Mulugo, L. et al. Unravelling technology-acceptance factors influencing farmer use of banana tissue culture planting materials in Central Uganda. African J. Sci. Technol. Innov. Dev. 12, 453–465 (2020). 57. Safi Sis, Y., Rezaei, A., Karimi, H. & Ataei, P. Modeling antecedent factors involved in behavioral intention towards technology application of genetically modified crops. GM Crops Food 13, 50–64 (2022). 58. Glover, D., Kim, S. K. & Stone, G. D. Golden Rice and technology adoption theory: A study of seed choice dynamics among rice growers in the Philippines. Technol. Soc. 60, 101227 (2020). 59. Ambali, O. I., Areal, F. J. & Georgantzis, N. Improved rice technology adoption: The role of spatially-dependent risk preference. Agriculture 11, 691 (2021). 60. Ayisi, D. N., Kozári, J. & Krisztina, T. Do smallholder farmers belong to the same adopter category? An assessment of smallholder farmers innovation adopter categories in Ghana. Heliyon 8, e10421 (2022). 61. Jamal, K., Kamarulzaman, N. H., Abdullah, A. M., Ismail, M. M. & Hashim, M. Adoption of fragrant rice farming: The case of paddy farmers in the East Coast Malaysia. UMK Procedia 1, 8–17 (2014). 62. Folorunso, O. & Ogunseye, S. O. Applying an enhanced technology acceptance model to knowledge management in agricultural extension services. Data Sci. J. 7, 31–45 (2008). 63. Marra, M. C. et al. Estimating the demand and willingness-to-pay for cotton yield monitors. Precis. Agric. 11, 215–238 (2010). 64. Hite, D., Hudson, D. & Intarapapong, W. Willingness to pay for water quality improvements: The case of precision application technology. J. Agric. Resour. Econ. 27, 433–449 (2002). 65. Hudson, D. & Hite, D. Producer willingness to pay for precision application technology: Implications for government and the technology industry. Can. J. Agric. Econ. Can. d’agroeconomie 51, 39–53 (2003). 66. Adrian, A. M., Norwood, S. H. & Mask, P. L. Producers’ perceptions and attitudes toward precision agriculture technologies. Comput. Electron. Agric. 48, 256–271 (2005). 67. Aubert, B. A., Schroeder, A. & Grimaudo, J. IT as enabler of sustainable farming: An empirical analysis of farmers’ adoption decision of precision agriculture technology. Decis. Support Syst. 54, 510–520 (2012). 68. Moore, G. C. & Benbasat, I. Development of an instrument to measure the perceptions of adopting an information technology innovation. Inf. Syst. Res. 2, 192–222 (1991). 69. Shiau, S., Huang, C.-Y., Yang, C.-L. & Juang, J.-N. A derivation of factors influencing the innovation diffusion of the OpenStreetMap in STEM education. Sustainability 10, 3447 (2018). 70. Ainembabazi, J. H. & Mugisha, J. The role of farming experience on the adoption of agricultural technologies: Evidence from smallholder farmers in Uganda. J. Dev. Stud. 50, 666–679 (2014). 71. Massresha, S. E., Lema, T. Z., Neway, M. M. & Degu, W. A. Perception and determinants of agricultural technology adoption in North Shoa Zone Amhara Regional State Ethiopia. Cogent Econ. Financ. https://doi.org/10.1080/23322039.2021.1956774 (2021). 72. Micha, E., Areal, F. J., Tranter, R. B. & Bailey, A. P. Uptake of agri-environmental schemes in the less-favoured areas of Greece: The role of corruption and farmers’ responses to the financial crisis. Land use policy 48, 144–157 (2015). 73. Saengavut, V. & Jirasatthumb, N. Smallholder decision-making process in technology adoption intention: implications for Dipterocarpus alatus in Northeastern Thailand. Heliyon 7, e06633 (2021). 74. Nyang’au, J. O., Mohamed, J. H., Mango, N., Makate, C. & Wangeci, A. N.,. Smallholder farmers’ perception of climate change and adoption of climate smart agriculture practices in Masaba South Sub-county, Kisii. Kenya. Heliyon 7, e06789 (2021). 75. Pierpaoli, E., Carli, G., Pignatti, E. & Canavari, M. Drivers of precision agriculture technologies adoption: A literature review. Procedia Technol. 8, 61–69 (2013). 76. Rizzo, G., Migliore, G., Schifani, G. & Vecchio, R. Key factors influencing farmers’ adoption of sustainable innovations: a systematic literature review and research agenda. Org. Agric. https://doi.org/10.1007/s13165-023-00440-7 (2023). 77. Knowler, D. & Bradshaw, B. Farmers’ adoption of conservation agriculture: A review and synthesis of recent research. Food Policy 32, 25–48 (2007). 78. Milkias, D. Determinants of agricultural technology adoption: the case of improved highland maize varieties in Toke Kutaye District, Oromia Regional State. Ethiopia. J. Invest. Manag. 7, 125 (2018). 79. Spielman, D., Lecoutere, E., Makhija, S. & Van Campenhout, B. Information and Communications Technology (ICT) and agricultural extension in developing countries. Annu. Rev. Resour. Econ. 13, 177–201 (2021). 80. Amankwah, A. Climate variability, agricultural technologies adoption, and productivity in rural Nigeria: a plot-level analysis. Agric. Food Secur. 12, 7 (2023). 81. Oyetunde-Usman, Z. Heterogenous Factors of Adoption of Agricultural Technologies in West and East Africa Countries: A Review. Front. Sustain. Food Syst. https://doi.org/10.3389/fsufs.2022.761498 (2022). 82. Grewal, R., Cote, J. A. & Baumgartner, H. Multicollinearity and Measurement Error in Structural Equation Models: Implications for Theory Testing. Mark. Sci. 23, 519–529 (2004). 83. Skiera, B., Reiner, J. & Albers, S. Regression Analysis. in Handbook of Market Research (eds. Homburg, C., Klarmann, M. & Vomberg, A.) 299–327 (Springer International Publishing, 2022). https://doi.org/10.1007/978-3-319-57413-4_17. 84. O’brien, R. M. A,. Caution regarding rules of thumb for variance inflation factors. Qual. Quant. 41, 673–690 (2007). 85. Ukoumunne, O. C., Gulliford, M. C. & Chinn, S. A note on the use of the variance inflation factor for determining sample size in cluster randomized trials. J. R. Stat. Soc. Ser. D (The Stat 51, 479–484 (2002). 86. Senaviratna, N. A. M. R., Cooray, A. & T. M. J.,. Diagnosing multicollinearity of logistic regression model. Asian J. Probab. Stat. https://doi.org/10.9734/ajpas/2019/v5i230132 (2019). 87. Yau, C. Y. & Hui, T. S. LARS-type algorithm for group lasso. Stat. Comput. 27, 1041–1048 (2017). 88. Di Mari, R., Rocci, R. & Gattone, S. A. LASSO-Penalized clusterwise linear regression modeling with Local Least Angle Regression (L-LARS). SSRN Electron. J. https://doi.org/10.2139/ssrn.3832769 (2021). 89. Sirimongkolkasem, T. & Drikvandi, R. On regularisation methods for analysis of high dimensional data. Ann. Data Sci. 6, 737– 763 (2019). 90. Freijeiro-González, L., Febrero-Bande, M. & González-Manteiga, W. A critical review of lasso and its derivatives for variable selection under dependence among covariates. Int. Stat. Rev. 90, 118–145 (2022). 91. Ranstam, J. & Cook, J. A. LASSO regression. Br. J. Surg. 105, 1348–1348 (2018). 92. Paul, S. & Drineas, P. Feature selection for ridge regression with provable guarantees. Neural Comput. 28, 716–742 (2015). 93. Ratner, B. Variable selection methods in regression: Ignorable problem, outing notable solution.. J. Target. Meas. Anal. Mark. 18, 65–75 (2010). 94. Chowdhury, M. Z. I. & Turin, T. C. Variable selection strategies and its importance in clinical prediction modelling. Fam. Med. Community Heal. 8, e000262 (2020). 95. Engelmann, B. Comprehensive Stepwise Selection for Logistic Regression. arXiv Prepr. 17 (2023) Scientific Reports | (2025) 15:8927 17| https://doi.org/10.1038/s41598-025-85173-1 www.nature.com/scientificreports/ https://doi.org/10.1080/23322039.2021.1956774 https://doi.org/10.1007/s13165-023-00440-7 https://doi.org/10.3389/fsufs.2022.761498 https://doi.org/10.1007/978-3-319-57413-4_17 https://doi.org/10.9734/ajpas/2019/v5i230132 https://doi.org/10.2139/ssrn.3832769 http://www.nature.com/scientificreports 96. Fotheringham, A. S., Brunsdon, C. & Charlton, M. 2002 Geographically Weighted Regression: The Analysis of Spatially Varying Relationships. International Journal of Geographical Information Science 28 (Taylor & Francis) 97. Chen, V. Y., Deng, W., Yang, T. & Matthews, S. A. Geographically Weighted Quantile Regression (GWQR): An Application to U.S Mortality Data. Geogr. Anal. 44, 134–150 (2012). 98. Hastie, T., Friedman, J. & Tibshirani, R. The Elements of Statistical Learning (Springer, New York, 2001). 99. Chicco, D. & Jurman, G. The Matthews correlation coefficient (MCC) should replace the ROC AUC as the standard metric for assessing binary classification. BioData Min. 16, 4 (2023). 100. Muschelli, J. ROC and AUC with a binary predictor: a potentially misleading metric. J. Classif. 37, 696–708 (2020). 101. Lobo, J. M., Jiménez-Valverde, A. & Real, R. AUC: a misleading measure of the performance of predictive distribution models. Glob. Ecol. Biogeogr. 17, 145–151 (2008). 102. He, Z., Li, L., Huang, Z. & Situ, H. Quantum-enhanced feature selection with forward selection and backward elimination. Quantum Inf. Process. 17, 154 (2018). 103. Staal, S. J., Baltenweck, I., Waithaka, M. M., DeWolff, T. & Njoroge, L. Location and uptake: integrated household and GIS analysis of technology adoption and land use, with application to smallholder dairy farms in Kenya. Agric. Econ. 27, 295–315 (2002). 104. Mishra, V. N., Kumar, V., Prasad, R. & Punia, M. Geographically weighted method integrated with logistic regression for analyzing spatially varying accuracy measures of remote sensing image classification. J. Indian Soc. Remote Sens. 49, 1189–1199 (2021). 105. Wang, Q., Jiang, D., Gao, Y., Zhang, Z. & Chang, Q. Examining the driving factors of SOM Using a Multi-Scale GWR model augmented by Geo-Detector and GWPCA Analysis. Agronomy 12, 1697 (2022). 106. Ruzzante, S., Labarta, R. & Bilton, A. Adoption of agricultural technology in the developing world: A meta-analysis of the empirical literature. World Dev. 146, 105599 (2021). 107. Qi, X., Liang, F., Yuan, W., Zhang, T. & Li, J. Factors influencing farmers’ adoption of eco-friendly fertilization technology in grain production: An integrated spatial–econometric analysis in China. J. Clean. Prod. 310, 127536 (2021). 108. Korir, L. et al. Adoption of dairy technologies in smallholder dairy farms in Ethiopia. Front. Sustain. Food Syst. ​h​t​t​p​s​:​/​/​d​o​i​.​o​r​g​/​1​0​ .​3​3​8​9​/​f​s​u​f​s​.​2​0​2​3​.​1​0​7​0​3​4​9​​​​ (2023). 109. Ghimire, R., Wen-chi, H. & Shrestha, R. B. Factors affecting adoption of improved rice varieties among rural farm households in central Nepal. Rice Sci. 22, 35–43 (2015). 110. Areal, F. J. & Pede, V. O. Modeling Spatial Interaction in Stochastic Frontier Analysis. Front. Sustain. Food Syst. ​h​t​t​p​s​:​/​/​d​o​i​.​o​r​g​/​1​0​ .​3​3​8​9​/​f​s​u​f​s​.​2​0​2​1​.​6​7​3​0​3​9​​​​ (2021). 111. Mamiit, R. J., Yanagida, J. & Miura, T. Productivity Hot Spots and Cold Spots: Setting Geographic Priorities for Achieving Food Production Targets. Front. Sustain. Food Syst. https://doi.org/10.3389/fsufs.2021.727484 (2021). 112. Houeninvo, G. H., Célestin Quenum, C. V. & Nonvide, G. M. A. Impact of improved maize variety adoption on smallholder farmers’ welfare in Benin. Econ. Innov. New Technol. 29, 831–846 (2020). 113. Genius, M., Koundouri, P., Nauges, C. & Tzouvelekas, V. Information Transmission in Irrigation Technology Adoption and Diffusion: Social Learning, Extension Services, and Spatial Effects. Am. J. Agric. Econ. 96, 328–344 (2014). 114. Wang, G., Lu, Q. & Capareda, S. C. Social network and extension service in farmers’ agricultural technology adoption efficiency. PLoS One 15, e0235927 (2020). 115. Bandiera, O. & Rasul, I. Social networks and technology adoption in northern mozambique. Econ. J. 116, 869–902 (2006). 116. Beaman, L. & Dillon, A. Diffusion of agricultural information within social networks: Evidence on gender inequalities from Mali. J. Dev. Econ. 133, 147–161 (2018). 117. Kassie, M., Jaleta, M., Shiferaw, B., Mmbando, F. & Mekuria, M. Adoption of interrelated sustainable agricultural practices in smallholder systems: Evidence from rural Tanzania. Technol. Forecast. Soc. Change 80, 525–540 (2013). 118. Ullah, A. et al. Role of local communities in forest landscape restoration: Key lessons from the billion trees afforestation project. Pakistan. Sci. Total Environ. 772, 145613 (2021). 119. Truong, V. D., Avula, R. Y., Pecota, K. V. & Yencho, G. C. Sweetpotato production, processing, and nutritional quality. Handb. Veg. Veg. Process. Second Ed. 2–2, 811–838 (2018). 120. Nguyen, L. V. et al. Variation in root growth responses of sweet potato to hypoxia and waterlogging. Vegetos 33, 367–375 (2020). 121. Ezin, V. et al. Assessment of production and marketing constraints and value chain of sweet potato in the municipalities of Dangbo and Bonou. Agric. Food Secur. 7, 15 (2018). 122. Collas, L. Decision-making in Agriculture: Why do farmers decide to adopt a new practice? (Brock University St Catharines, 2018). 123. Dong, H., Wang, H. & Han, J. Understanding ecological agricultural technology adoption in china using an integrated technology acceptance model—theory of planned behavior model. Front. Environ. Sci. 10, 927668 (2022). 124. Jimenez, I. A. C., García, L. C. C., Marcolin, F., Violante, M. G. & Vezzetti, E. Validation of a TAM Extension in Agriculture: Exploring the Determinants of Acceptance of an e-Learning Platform. Appl. Sci. 11(10), 4672 (2021). 125. Dadzie, S. K. N., Ndebugri, J., Inkoom, E. W. & Akuamoah-Boateng, S. Social networking and risk attitudes nexus: implication for technology adoption among smallholder cassava farmers in Ghana. Agric. Food Secur. 11, 41 (2022). 126. Bessant, J., Öberg, C. & Trifilova, A. Framing problems in radical innovation. Ind. Mark. Manag. 43, 1284–1292 (2014). 127. Cakirli Akyüz, N. & Theuvsen, L. The impact of behavioral drivers on adoption of sustainable agricultural practices: The case of organic farming in Turkey. Sustainability 12, 6875 (2020). 128. Munasib, A., Roy, D. & Birol, E. Networks and low adoption of modern technology: The case of pearl millet in Rajasthan India. Indian Grow. & Dev. Rev. https://doi.org/10.1108/IGDR-07-2014-0025 (2015). 129. Rosário, J., Madureira, L., Marques, C. & Silva, R. Understanding farmers’ adoption of sustainable agriculture innovations: A systematic literature review. Agronomy 12, 2879 (2022). 130. Mudege, N. N., Mayanja, S. & Muzhingi, T. Women and men farmer perceptions of economic and health benefits of orange fleshed sweet potato (OFSP) in Phalombe and Chikwawa districts in Malawi. Food Secur. 9, 387–400 (2017). 131. Zawedde, B. M., Harris, C., Alajo, A., Hancock, J. & Grumet, R. Factors influencing diversity of farmers’ varieties of sweet potato in Uganda: implications for conservation. Econ. Bot. 68, 337–349 (2014). 132. Akoègninou, A., Van der Burg, W. J. & Van der Maesen, L. J. G. Flore analytique du Bénin (Backhuys, 2006). 133. Rodriguez Izaba, O. F., Torres, A. P., Marshall, M. I. & Thompson, A. W. Market access and value-added strategies in the specialty crops industry. HortScience 58, 32–39 (2023). 134. Heindorf, C., Reyes-Agüero, J. A. & van’t Hooft, A.,. Local markets: agrobiodiversity reservoirs and access points for farmers’ plant propagation materials. Front. Sustain. Food Syst. 5, 597822 (2021). 135. Ahoudou, I. et al. Farmers’ selection criteria for sweet potato varieties in Benin: An application of best-worst scaling. Exp. Agric. 59, e25 (2023). 136. Tobler, W. R. A computer movie simulating urban growth in the detroit region. Econ. Geogr. 46, 234 (1970). Acknowledgements We express sincere gratitude for the financial support received from the African Union through the project titled 'Improved OFSP Value Chain for Food and Nutrition Security in Benin, Niger, and Nigeria’ (No. AUG II- 2-247-2018). Our heartfelt appreciation goes to the municipality mayors and village heads for their unwavering dedication and assistance throughout the project. Special recognition is extended to our diligent field enumer- Scientific Reports | (2025) 15:8927 18| https://doi.org/10.1038/s41598-025-85173-1 www.nature.com/scientificreports/ https://doi.org/10.3389/fsufs.2023.1070349 https://doi.org/10.3389/fsufs.2023.1070349 https://doi.org/10.3389/fsufs.2021.673039 https://doi.org/10.3389/fsufs.2021.673039 https://doi.org/10.3389/fsufs.2021.727484 https://doi.org/10.1108/IGDR-07-2014-0025 http://www.nature.com/scientificreports ators Hadid A. Gangni-Ahossou, Hervé M. D. Codja, Lionel K. H. Guedou, Mses. Olivia V. Dadesso, Nadège Y. Dossou of the Faculty of Agronomic Sciences, University of Abomey-Calavi whose efforts and hard work have been invaluable. Author contributions I.A.: Conceptualization, Methodology, Formal analysis, Investigation, Data Curation, Writing—Original Draft, Writing—Review & Editing, Visualization. N.V.F.H.: Writing—Review & Editing, Project administration. C.O.A.A.: Writing—Review & Editing. I.T.L.A.: Investigation, Resources. D.E.O.S.: Writing—Review & Edit- ing, Visualization, Supervision. N.F.F.: Writing—Review & Editing. F.A.K.: Writing—Review & Editing. I.M.M: Writing—Review & Editing, Funding acquisition, Project administration. E.G.A.D.: Conceptualization, Super- vision, Writing—Review & Editing, Funding acquisition, Project administration. Funding African Union Commission, Grant/Award (AUG (II-2-247-2018)). Declarations Competing interests The authors declare no competing interests. Ethics approval The aims of the study were thoroughly communicated to both local authorities and participants. Prior to distributing the questionnaire, we sought individual consent from the authorities and respondents. Participation in the study was strictly voluntary, and only those who expressed their willingness to participate were included. We upheld the confidentiality of the participants by ensuring their identities were anonymized in our databases. Approval for human experiments Before conducting the interview, verbal informed consent was obtained from each participant, ensuring they were fully informed of our objectives. Additional information Supplementary Information The online version contains supplementary material available at ​h​t​t​p​s​:​/​/​d​o​i​.​o​r​g​/​1​ 0​.​1​0​3​8​/​s​4​1​5​9​8​-​0​2​5​-​8​5​1​7​3​-​1​​​​​.​​ Correspondence and requests for materials should be addressed to I.A. or E.G.A.-D. Reprints and permissions information is available at www.nature.com/reprints. Publisher’s note  Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access  This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit ​h​t​t​p​:​/​/​c​r​e​a​t​i​v​e​c​o​m​m​o​ n​s​.​o​r​g​/​l​i​c​e​n​s​e​s​/​b​y​-​n​c​-​n​d​/​4​.​0​/​​​​​.​​ © The Author(s) 2025 Scientific Reports | (2025) 15:8927 19| https://doi.org/10.1038/s41598-025-85173-1 www.nature.com/scientificreports/ https://doi.org/10.1038/s41598-025-85173-1 https://doi.org/10.1038/s41598-025-85173-1 http://creativecommons.org/licenses/by-nc-nd/4.0/ http://creativecommons.org/licenses/by-nc-nd/4.0/ http://www.nature.com/scientificreports Evaluating logistic regression and geographically weighted logistic regression models for predicting orange-fleshed sweet potato adoption intention in Benin Conceptual frameworks and hypothesis Technology acceptance model (TAM) Innovation diffusion theory (IDT) and perceived characteristics of innovation (PCI) Integration of technology acceptance model (TAM) and perceived characteristics of innovation (PCI) External factors effect on new technology adoption by farmers Socio-demographic characteristics Farm characteristics and geographical context Hypothesis Material and methods Study area and sampling method Data collection Data analysis Variables selection Logistic regression model Geographically weighted logistic regression model Models evaluation Model validation Results Farmers socio-economic characteristics Variable selection and logistic regression parameter estimation Geographically weighted logistic regression parameter estimation Model evaluation Prediction of OFSP adoption probability in Benin Discussion Conclusion Study limitation References