Farm Households Food Security Status Automation Through Supervised Learning Approach: A Look at Agroecological Farms

Abstract

Food insecurity is a pervasive phenomenon in Africa, and the paradox is that it affects farming households more than others. Although early and accurate detection of famine in farm households is a complex challenge, it remains very important to help reduce their vulnerability and inform policy and practice. Existing food security monitoring tools focus on isolated dimensions of the problem and some remain static. In this study, we focus on a composite metric of food security which combines three key indicators: Food Consumption Score (FCS), Livelihood Coping Strategies (LCS), and Food Expenditure Share (FES). We trained and tested various automatic classifications models, applied them to Burkina Faso's 2020 Permanent Agricultural Survey data to predict food security and validated our results with data from 2019. This prediction was compared to actual food security. The approach used correctly identifies the food security status of 84% of the households. Farms which practice agroecology are slightly less affected by food insecurity than those practicing conventional agriculture. We also find that households that implement integrated livestock-agriculture systems are less affected by food insecurity than households that do not. The proposed models can speedily predict food security status using multidimensional datasets and are able to identify the risk factors associated with predicted food insecurity trends, which is key to target priority areas for intervention by policymakers. They could also be integrated into data collection tools via server communication protocols to enable real-time monitoring.