Abstract:
Agroecology promotion is nowadays an important objective recognized almost everywhere because of the perspectives it offers in terms of sustainability. However, despite the initiatives of numerous actors around the world, agroecology's adoption rate remains low. In this study, we combine multidimensional analysis methods with machine learning methods to establish a typology of farms in Burkina Faso using data from the permanent agricultural survey covering the whole country. Multidimensional analysis methods were used to label the data, and the classification algorithms were trained on these labels. Our results reveal three major farm classes: conventional farms, farms in agroecological transition, and agroecological farms. The parametric models performed better than the non-parametric models (Kappa of 99% versus 50%). The good performance of the radial support vector machine model reveals the existence of a non-linear relationship in some of the study's features, and this could not have been highlighted by traditional multivariate analysis alone. Algorithms that have performed well can, for example, be integrated into agricultural data collection applications to inform decision-makers about the adoption of agroecology in real time and suggest solutions to improve the dimensions in which farms are weak. Agroecological farms make up around 9%, (95% CI=[8−11%]) of all farms, compared with 21%, (95% CI=[19−24%]) for conventional farms. The geographical distribution shows that agroecology appears to be strongly supported by regions that face significant environmental and climatic constraints. Marital status, gender, membership in farmers' organizations, access to agricultural credit, and labor cost are identified as levers of transition.