A team made up of researchers from several countries has developed a set of machine learning models which ensure that they will be able, in the short term, predict food shortages. Thanks to this, they will be able to help governments and various international agencies and entities to understand where they can best help.
Development scientists are members of the world food programof Department of mathematics at the University of London and of Central European University Data Science and Network Department. They all used a “global single dataset» for the development of machine learning models capable of explaining up to 81% of the variations in insufficient food consumption.
The machine learning models developed are based on indirect data sources from areas such as food prices, macroeconomic indicators (including GDP), weather, conflict, the prevalence of malnutrition, and other previous food insecurity trends. The intention behind all this is that, with the help of the models, short-term forecasts can be created. The outputs from the models have been used to create a world map showing short-term food insecurity forecasts, called HungerMap (Map of hunger).
According to the article published in Nature Food on these models, and on the research itself, the team that has carried out the research work shows «that the proposed models can offer a short-term forecast of the food security situation, almost in real time, and propose a method to identify which variables are driving the observed changes in the expected trends, something key for the predictions to serve those who they make the decisions«.
In the article, the developers of the models also note that “Food insecurity is a more dynamic and unstable phenomenon than poverty, with a seasonal component related to agricultural production calendars, and subject to rapid changes when external shocks impact, requiring frequent and rapid assessments. This opens the door to short-term prediction of food security on a global scale, allowing decision makers to do so in a more accurate and informed manner in relation to policies and programs aimed at fighting hunger.«.
Researchers have also used secondary data to predict long-term food insecurity. They have done it with information such as agricultural production, climate models and crop statistics. However, they have made projections up to 2030 on changes in crop production.
The models have been developedBased on a specific WFP need, they fill a gap due to inaccessibility and limited resources to provide information on a regular basis from the least accessible places, where food security assessments are carried out only once or twice a year , and who need a constant flow of information to communicate humanitarian operations«.
Its authors point out that when their models predict increases in the prevalence of food insecurity, then the World Food Program could activate rapid assessments, face-to-face or remote surveys, as well as mobilize in-country analytics to better understand the situation. .
