Forecasting food insecurity with machine learning using data from the WFP

17 Mar 2023 --- Scientists are presenting a tool that can forecast food insecurity 30 days in advance for countries with insufficient food consumption using the eXtreme Gradient Boosting (XGBoost) algorithm – a machine learning technique using several models at once for a better output. The researchers, from the Central European University, Vienna, Austria, use data from the World Food Programme (WFP) and argue that the tool might help prevent food insecurity. 

“From a policy perspective, food insecurity will remain a worldwide concern for the next 50 years and beyond,” reads the study. The authors examine food insecurity in six countries across the Middle East and West Africa – Syria, Yemen, Nigeria, Mali, Cameroon and Burkina Faso, all experiencing food insecurity driven by conflict, extreme weather and economic shocks.

Published in Scientific Reports, the researchers aim to forecast the daily evolution of the prevalence of people with insufficient food consumption below the national level. Using the Food Consumption Score (FCS) – an indicator capturing household dietary diversity and nutritional intake. 

Access to predictions of food insecurity allows governments and organizations to identify which areas should be monitored more closely and eventually make timely decisions on resource allocation.

The study authors point out that conflict can be both the cause or the consequence of food insecurity.Conflicts and extreme weather
Food insecurity is characterized by four main factors: availability, access, utilization and stability. It results from sociodemographic, environmental and political events, with climate variabilities and extreme weather characterized as the main driver, as it affects crop production, child nutrition and overall food insecurity. 

The study authors point out that conflict can be both the cause or the consequence of food insecurity, as it impacts agricultural production and nutritional intake.

The forecasting model aims to predict the daily sub-national prevalence of food insecurity in countries experiencing major food crises using a time series from the WFP.

Exploring if secondary information can be used to predict future evolution and food insecurity 30 days ahead, the study found that the used model made it possible “with higher accuracy than a naive approach solely based on the last measured prevalence, at least in places where enough training data are available to inform the model,” notes the study.

For a timely response
The study took Ramadan, extreme weather and conflict into account and stressed that it was conducted during the COVID-19 pandemic, which affected all the mentioned countries while exacerbating food insecurity on a global scale. 

The authors argue that the tool can help decision makers to have advanced access to information in areas most at risk of food insecurity situations and that it would allow for a more timely response. 

“Predictions should be used with caution and considered only as an indication of what may happen in the near future, hence informing preparedness efforts by suggesting a need for further in-depth assessments of the food security situation,” the study notes.

Recently, the African Bank Development Group hosted an event to serve as a “wake-up call” for African nations, as political actions must step it up to reach the nutritional targets set for 2025.

Edited by Beatrice Wihlander

To contact our editorial team please email us at editorial@cnsmedia.com