Exploring the determinants of biopesticide adoption among farmers using machine learning: a study from Bihar, India

The adoption of sustainable agricultural practices remains a critical challenge in developing countries. This study moves beyond traditional linear models to employ a multi-faceted machine learning (ML) approach, aiming to predict biopesticide adoption and to uncover the most influential behavioural constructs of farmers towards biopesticide adoption intention in Bihar, India. The data were collected from 468 farmers across four districts of Bihar, using a structured questionnaire to measure behavioural constructs. An analytical framework was deployed by using seven supervised ML algorithms (including Random Forest, Support Vector Machine, Logistic Regression, K-nearest neighbors, naive Bayes, Neural Network and Decision Tree), which were trained and rigorously tested to predict adoption behaviour. The Random Forest model demonstrated the highest predictive efficacy, with an accuracy of 78.7% and a robust Kappa score of 0.571. A consistent finding across all models reported the importance of subjective norms and attitude as the primary factors associated with adoption, while perceived benefits were placed in the last position. This study provides a data-driven blueprint for designing targeted interventions. By leveraging social networks to influence subjective norms and increasing communication to address the specific attitudinal and practical barriers of farmers, the government and policymakers. This can create more effective strategies to accelerate the biopesticide adoption for sustainable agricultural practices.