Using AI to unveil smallholder farmers’ crop selection preferences in rural Ethiopia

Abstract This study contributes to the existing literature by empirically examining the factors shaping farmers’ decisions on crop selection in Ethiopia using machine learning methods. In this context, we use data from rural Ethiopian communities, including hydrological, social, economic, and demographic variables. Our approach leverages advanced machine learning methods to pinpoint the key parameters shaping these decisions, drawing on the observed effects of a citizen science program and identifying socially valuable traits linked to water management. The results show that farming decisions are based on an amalgamation of hydrological, social, and economic parameters, challenging domain-specific grounded approaches. The ensemble machine learning model demonstrated superior performance in comparison to independent models in forecasting crop selection in terms of both balanced accuracy and F 1 score metrics. Key findings from our analysis show that the most influential variables across all models are the type of seed and the planting month. In rainfed regions, rainfall-related features and social factors, specifically leadership in community organizations, also appear highly important. In contrast, in irrigated fields, the use of pesticides during periods of water shortage emerges as a high-impact and unique variable. These results provide a more nuanced understanding of the factors that influence farmers’ choices and offer insights into shaping agricultural policies.