Determinants of adoption of climate-smart agriculture (CSA) practices in mushroom farming in Bangladesh.

Climate-Smart Agriculture (CSA) practices can reduce effects of climate change in agriculture by increasing production efficiency. Yet, adoption of CSA practices is not universal on Bangladesh farms. This study aim is to identify determinants of adoption of CSA practices in mushroom farming in Bangladesh using a combination of frequentist approaches and machine learning (ML). A total 150 mushroom farmers were selected from Savar upazila. Farmers were interviewed using a questionnaire, and results were analyzed using a combination of Bayesian and ML approaches. Among respondents, 48% of farmers had adopted at least one CSA practice for mushroom farming. Bayesian analysis revealed that mushroom farmers with secondary education, prior knowledge of CSA, training related to mushroom cultivation, access to climate information and credit were more likely to adopt CSA practice for mushroom production compared to their peers. Additionally, new farmers had higher odds of adopting CSA than their counterparts. In terms of predicting adoption of CSA practices using ML, the support vector machine algorithm slightly outperformed other ML algorithms, with an estimated accuracy of 87.3%, recall of 92%, F-score of 87.9%, g-mean score of 87.7%, Cohen’s Kappa of 74.6%, and Matthews’ correlation coefficient of 74.7%. However, for area under the curve and precision recall curve, gradient boosting machine showed better performance. Consistent with the frequentist approach, prior knowledge of CSA practices was among the most influent factors towards adoption of CSA practices for mushroom farming, followed by access to climate information, farm ownership, training related to mushroom, access to credit and internet. We hypothesize that targeted training, and access to climate information and credit will increase adoption of CSA practices in mushroom farming in Bangladesh.