Des chercheurs créent des modèles de ML pour prévoir les pénuries alimentaires

Une équipe internationale de chercheurs a construit un ensemble de modèles d'apprentissage automatique qui, selon eux, peuvent aider à prévoir les pénuries alimentaires mondiales dans un avenir proche, aidant les gouvernements et les agences internationales à comprendre où ils peuvent le mieux aider.

Scientists from the World Food Programme, University of London Mathematics Department and Central European University Department of Network and Data Science, made use of a "unique global dataset" to build machine learning models that can explain up to 81 percent of the variation in insufficient food consumption.

The study claims the machine learning models draw from indirect data sources in areas such as food prices, macro-economic indicators (including GDP), weather, conflict, prevalence of undernourishment, population density, and previous food insecurity trends. The aim is to create near-term forecasts, or "nowcasts."

"We show that the proposed models can nowcast the food security situation in near real-time and propose a method to identify which variables are driving the changes observed in predicted trends — which is key to make predictions serviceable to decision-makers," the research paper publié dans Nature Food cette semaine .

Les sorties des modèles ML ont été utilisées pour créer une carte du monde comprenant des prévisions d'insécurité alimentaire à court terme appelées Carte de la faim.

In 2019, the number of undernourished people was estimated to be 650 million, with 135 million in 55 countries and territories reported to be acutely "food-insecure." Food insecure is defined as lacking consistent access to enough food so you can lead an active, healthy life. Following the global COVID-19 pandemic, these numbers shot up. At least 280 million people were reported to be acutely food insecure in 2020, more than doubling the number from the previous year.

Governments and international organizations such as the World Food Programme (WFP), the Food and Agriculture Organization (FAO) and the World Bank measure food security with face-to-face surveys or remote mobile phone surveys. But these can be costly, while accuracy can be a problem. "Food insecurity is a more dynamic and unstable phenomenon than poverty, with a seasonal component related to agricultural production calendars and subject to swift changes when external shocks hit, therefore requiring more frequent and rapid assessments," the paper said.

"This opens the door to food security near-real time nowcasting on a global scale, allowing decision-makers to make more timely and informed decisions on policies and programmes oriented towards the fight against hunger," the authors said.

Les chercheurs ont également utilisé des données secondaires pour prédire l'insécurité alimentaire à plus long terme. La production agricole, les modèles statistiques de cultures et la modélisation climatique ont été utilisés pour faire des projections jusqu'en 2030 sur les changements de la production agricole. Pendant ce temps, des données de téléphonie mobile anonymisées et agrégées ont été utilisées au Sénégal pour examiner le mouvement plus large des personnes au fil des saisons, et ont été combinées avec des calendriers agricoles et des enregistrements de précipitations pour caractériser la sécurité alimentaire.

The current study did not use mobile phone data because it is typically obtained through national mobile phone operators. "Therefore, it is not a type of data that is easily scalable, and this is why it was not a suitable data source for our global approach," lead author Elisa Omodei, assistant professor at the Central European University told Le registre.

Les auteurs suggèrent que lorsque leurs modèles prédisent une augmentation de la prévalence des personnes en situation d'insécurité alimentaire, le Programme alimentaire mondial déclencherait des évaluations rapides par le biais d'enquêtes en face à face ou à distance et mobiliserait des analystes dans le pays pour mieux comprendre la situation.

"The development of these models is motivated by a specific need of WFP to fill a gap that currently exists because of limited resources and inaccessibility, that is, to provide regular information for less reachable places, where food security assessments are carried out only once or twice per year but that nevertheless require a constant flow of information to inform humanitarian operations," the paper said. ®