AI system can predict child malnutrition before it strikes

Kenya is grappling with a child malnutrition crisis. Nearly 350,000 children under five suffer from acute malnutrition, a condition that severely weakens their immune systems. Some regions report malnutrition rates as high as 25 percent.

Now, a team from University of Southern California (USC), Microsoft AI for Good Lab, Amref Health Africa, and Kenya’s Ministry of Health has developed an artificial intelligence (AI) model that can predict malnutrition up to six months ahead.

AI predicts child malnutrition

The model combines clinical records from 17,000 health facilities with satellite data on crop health. The goal is to identify where malnutrition is likely to spike next.

Unlike previous methods, which focus mainly on historical malnutrition rates, this AI-driven model uses complex data patterns.

Bistra Dilkina, co-director of USC’s Center for Artificial Intelligence in Society, said the model is a “game-changer.”

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

Child malnutrition trends

The data comes from Kenya’s District Health Information System 2 (DHIS2), a platform that collects health data from clinics nationwide.

In addition, satellite imagery from NASA’s Moderate Resolution Imaging Spectroradiometer (MODIS) measures crop productivity. Gross Primary Productivity (GPP) indicates how well crops are growing, acting as a proxy for food security.

Regions with poor crop health often show higher malnutrition rates. By analyzing both health and satellite data, the model predicts where malnutrition might rise. It achieved 89% accuracy in one-month forecasts and maintained 86% accuracy over six months – a notable leap from older models.

In Kenya, 5% of children under five suffer from acute malnutrition. Globally, undernutrition contributes to nearly half of all deaths in this age group. Laura Ferguson, director of research at USC’s Institute on Inequalities in Global Health, emphasized the urgency.

“Malnutrition is a public health emergency in Kenya,” said Ferguson. “Children are sick unnecessarily. Children are dying unnecessarily.”

Satellite data and malnutrition forecasts

Traditional forecasting relies on expert judgment and past trends. But in regions where malnutrition rates fluctuate, this approach often falls short. The new AI model uses data-driven insights to fill these gaps.

Murage S.M. Kiongo, Program Officer for Monitoring and Evaluation at Kenya’s Ministry of Health, explained the model’s potential.

“The best way to predict the future is to create it using available data for better planning and prepositioning in developing countries,” said Kiongo.

Which forecasting model works best?

The research team tested three forecasting methods: Window Averaging (WA), Logistic Regression (LR), and Gradient Boosting (GB).

The GB model led the pack, reaching 86% accuracy over six months. WA lagged with a 73% accuracy, proving that simple historical averages can’t capture complex data patterns.

Interestingly, GPP data alone performed almost as well as clinical data. In regions where health facility data is scarce, satellite imagery could serve as a crucial tool.

Pinpointing malnutrition risk areas

Not every region faces the same risk. The model highlighted areas like Turkana and Kuria West, where malnutrition rates exceed 15%. These hotspots often suffer from poor crop yields and limited access to healthcare.

By pinpointing these areas, the model helps humanitarian organizations and government agencies intervene sooner, potentially saving lives.

To make these predictions actionable, the team built a dashboard. It visualizes malnutrition risks across Kenya, integrating clinical data, GPP readings, and forecasting outcomes.

The dashboard allows policymakers to see where malnutrition might spike next, enabling faster, targeted responses.

Actionable child malnutrition insights

Laura Ferguson and Bistra Dilkina are working with Kenya’s Ministry of Health and Amref Health Africa to integrate the dashboard into national systems. The goal is to make this tool a regular part of decision-making, ensuring resources reach those most in need.

Data gaps remain a challenge. Many rural children never visit health clinics, leaving them out of the DHIS2 dataset. Reporting inconsistencies also pose problems, as do mismatched administrative boundaries.

The research team plans to address these gaps by adding more data sources, like rainfall patterns and crop yields. They’re also exploring ways to adapt the model for other countries, particularly those using DHIS2.

A global framework

The AI model has far-reaching implications. Over 125 countries use DHIS2, including 80 where child malnutrition is rampant. If the model can predict malnutrition in Kenya, it can likely do the same elsewhere.

The study, titled “Forecasting acute childhood malnutrition in Kenya using machine learning and diverse sets of indicators,” was published in PLOS One on May 14, 2025. The authors include Girmaw Abebe Tadesse from Microsoft AI for Good Lab, Laura Ferguson from USC, and Bistra Dilkina.

“If we can do this for Kenya, we can do it for other countries,” said Dilkina. “The sky’s the limit when there is a genuine commitment to work in partnerships.”

Early warnings of child malnutrition

In regions where food insecurity and malnutrition are intertwined, early warnings can mean the difference between life and death. By predicting malnutrition months in advance, the AI model provides a crucial window for action.

For humanitarian organizations and governments, this tool offers more than predictions. It provides a plan – one based on data, not estimates. In a country where child malnutrition claims thousands of young lives each year, that plan could be a lifesaver.

Now, the focus shifts to scaling this model. Kenya is just the start. With the right data and partnerships, the AI model could become a global framework, predicting malnutrition and preventing suffering in countries where children are most at risk.

The study is published in the journal PLOS One.

—–

Like what you read? Subscribe to our newsletter for engaging articles, exclusive content, and the latest updates. 

Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.

—–