What is it about?
In 2024, about two in three Nigerians lived in households facing moderate or severe food insecurity, according to the Nigeria Demographic and Health Survey (NDHS). That survey round was the first to ask households directly about their access to food. This study asks whether free, openly available information can show which households are most at risk. It combines survey answers from nearly 40,000 households with five map and satellite sources: rainfall, vegetation, nearby conflict events, population density, and travel time to the nearest city. Computer models were trained on part of the data and then tested on households from communities they had never seen. In about 73 out of 100 comparisons between a food-insecure and a food-secure household, the best model gave the food-insecure household the higher risk score. Where a household lives turned out to matter about as much as its own circumstances, such as wealth and household size.
Featured Image
Photo by Omotayo Tajudeen on Unsplash
Why is it important?
Food prices in Nigeria rose sharply after the May 2023 removal of fuel subsidies and the June 2023 naira devaluation. Programmes that send cash or food to poor households need to know where to look, and most targeting today relies on household characteristics alone. This study shows that adding free geographic information improves the picture. It also releases the full method as open code, so governments, researchers and agencies can check it or reuse it. The study also carries two cautions. First, testing a model on households from the same communities it learned from makes it look better than it is; the study measures that gap and shows how to avoid it. Second, when the models had to predict for a whole region of Nigeria they had never seen, a simple statistical model did better than the more complex ones, so local data still matter before any model is used in a new area. One finding needs further study: households farther from cities were less likely to be food insecure, possibly because they grow more of their own food.
Perspectives
I work at the National Population Commission, which carried out the 2024 NDHS fieldwork, so I have seen how much effort goes into every household interview. I wanted that data to do more work after the report was published. The first version of this paper tested its models the usual way, on a random set of households. The reviewers pointed out that households in the same community share the same location data, so the models were partly being tested on places they had already seen. Fixing that lowered the headline accuracy for one model from 0.77 to about 0.71, and it made the paper more honest. I now think this check should be routine for anyone applying machine learning to cluster surveys like the DHS. If you work on food security, social protection or survey data, in Nigeria or elsewhere, I would be glad to hear from you.
Philip Osung Osung
Nile University of Nigeria, Abuja
Read the Original
This page is a summary of: Household-level food-insecurity prediction from experience-scale surveys and open geospatial data: A reproducible machine-learning pipeline for Nigeria, PLOS One, October 2026, PLOS,
DOI: 10.1371/journal.pone.0349978.
You can read the full text:
Resources
Contributors
The following have contributed to this page
