[NAIROBI, SciDev.Net] An Artificial Intelligence (AI) tool that predicts acute child malnutrition up to six months in advance could help combat the condition in Kenya and across Africa, researchers say.

Malnutrition is one of the leading causes of mortality globally, with nearly half of deaths among children under five linked to acute undernutrition – most of them in low- and middle-income countries – according to the World Health Organization.

However, gaps in data can make it difficult to know where to focus resources in countries like Kenya.

Five per cent of children in Kenya are acutely malnourished, according to the 2022 Kenya Demographic Health Survey, a level considered a public health concern.

Scientists have come up with a machine learning model that uses clinical health data and satellite imagery to forecast malnutrition trends across the country.

The tool was developed by a team from the University of Southern California (USC), in collaboration with Microsoft’s AI for Good Research Lab, Amref Health Africa, and Kenya’s Ministry of Health.

Lead researcher Laura Ferguson, director of research at the USC Institute on Inequalities in Global Health, says the goal is to equip health authorities with early warnings that support effective prevention and treatment responses.

“The tool is designed to predict malnutrition across counties in Kenya [and]… prepare prevention and treatment strategies,” Ferguson told SciDev.Net.

To make these forecasts, the model pulls data from the government’s District Health Information Software System (DHIS2) and combines it with satellite imagery to pinpoint where and when malnutrition is likely to occur.

Unlike traditional models that depend solely on historical trends, this AI tool integrates clinical data from more than 17,000 Kenyan health facilities.

It achieved 89 per cent accuracy for one-month predictions and 86 per cent accuracy over six months, marking a significant improvement over baseline models.

The tool can also integrate publicly available data on agricultural vegetation derived from satellite imagery into the model, to indicate available food sources, Ferguson added.

Encouraged by the results in Kenya, the researchers hope the tool can be adapted for use in nearly 125 other countries that also use DHIS2 — particularly in the 80 low- and middle-income nations where malnutrition remains a leading cause of child mortality.

“This model is a game-changer,” said Bistra Dilkina, associate professor of computer science and co-director of the USC Center for AI in Society.

“By using data-driven AI models, you can capture more complex relationships between multiple variables that work together to help us predict malnutrition more accurately,” she explained.

To maximise the impact of the tool, collaboration across sectors is key, says Samuel Mburu, head of digital transformation at Amref Health Africa, who also worked on the project. He suggests aligning health services with agriculture and disaster management efforts.

“Continued investment in digital health infrastructure and training is also critical,” Mburu told SciDev.Net.

Peter Ofware, Kenya country director for Helen Keller International, a US-based non-profit focused on nutrition and health, agrees that integrating vegetation data with DHIS2 improves forecasting accuracy.

“This improves the accuracy of forecasts,” said Ofware, who did not participate in the research.

“However, DHIS data, which is their primary source, has many limitations in quality —especially for malnutrition.”

Children are typically only screened for malnutrition in facilities where treatment is available, which limits how representative the data is, he added.

This article was produced by SciDev.Net’s Sub-Saharan Africa desk.