Basis for the application of machine learning in monitoring and anticipating food crises in Central America

Abstract

The article offers a detailed and updated review on the application of data science tools based on machine learning algorithms in order to predict the short and medium term probability of food crises in territories of countries with high vulnerability to this type of situation. After a brief review of the definition of food security and its metrics, the main international efforts are described to monitor the agroclimatic, economic and sociopolitical factors that most affect the nutritional deterioration of population groups or specific geographic areas, and then generate alerts that trigger humanitarian assistance to prevent the increase in hunger and its effects on the health of those who suffer from it. Based on the review carried out, a prediction model adapted to the context of the Central American countries is proposed, in which structural variables are considered to be used in the annual determination of food vulnerability profiles, as well as others subject to permanent changes and that therefore allow the identification of shocks or disturbances that can impact food security. The proposed model seeks to improve decision-making and prioritization of resources and humanitarian assistance in regions with limited data availability.