Globally, nearly half of the deaths of children under five are linked to malnutrition, and in Kenya, it’s the leading cause of illness and death among this age group. But now, scientists have developed a machine-learning model to forecast acute cases among youngsters in this country.
Laura Ferguson and Bistra Dilkina write in The Conversation:
Children with malnutrition typically show signs of recent and severe weight loss, and may also have swollen ankles and feet. It is usually the result of insufficient food or having infectious diseases, especially diarrhoea.
Acute malnutrition weakens a child’s immune system, which can lead to increased susceptibility to infectious diseases like pneumonia. It can also cause more severe illness and an increased risk of death.
Currently, the Kenyan national response to malnutrition, implemented by the Ministry of Health, is based on historical trends of malnutrition – so when cases have been reported in a certain month, the Ministry anticipates a repeat during a similar month in subsequent years.
No statistical modelling guides responses, which has limited their accuracy.
The Ministry has collected monthly data on nutrition-related indicators and other health conditions for many years, and our multi-disciplinary team set out to explore whether we could use these data to help forecast where, geographically, child malnutrition was likely to occur in the near future. We were aiming for a more accurate forecast than the existing method.
We developed a machine-learning model to forecast acute malnutrition among children in the country – a type of mathematical model that, once “trained” on an existing data set, can make predictions of future outcomes. We used existing data and improved forecasting capabilities by including complementary data sources, such as satellite imagery that provides an indicator of crop health.
We found that machine learning-based models consistently outperformed existing platforms used to forecast malnutrition rates in Kenya. And we found that models with satellite-based features worked even better.
Our results, published in PLOS One, demonstrate the ability of machine learning models to more accurately forecast malnutrition in Kenya up to six months before, from a variety of indicators.
If we have advance knowledge of where malnutrition is likely to be high, scarce resources can be allocated to these high-risk areas in a timely manner to try to prevent children from becoming malnourished.
How we did it
We used clinical data from the Kenya Health Information System, which included data on diarrhoea treatment and low birth weight. We collected data on children who visited a health facility who met the definition of being acutely malnourished, among other relevant clinical indicators.
Given that food insecurity is a key driver, we also incorporated data reflecting crop activity into our models. We used a NASA satellite to look at gross primary productivity, which measures the rate at which plants convert solar energy into chemical energy. This provides a coarse indicator of crop health and productivity.
Lower average rates can be an early indication of food scarcity.
We tested several methods and models for forecasting malnutrition risk using data collected from January 2019 to February 2024. The gradient boosting machine learning model – trained on previous acute malnutrition outcomes and gross primary productivity measurements – turned out to be the most effective model for forecasting acute malnutrition among children.
This model can forecast where and at what prevalence level acute malnutrition is likely to occur in one month’s time with 89% accuracy.
All of the models we developed performed well where the prevalence of acute child malnutrition was expected to be at more than 30%, for instance in northern and eastern Kenya, which have dry climates. However, when the prevalence was less than 15%, for instance in western and central Kenya, only the machine learning models were able to forecast with good accuracy.
This higher accuracy is achieved because the models use additional information on multiple clinical factors. They can, therefore, find more complex relationships.
Implications
Current efforts to predict acute malnutrition among children rely only on historical knowledge of malnutrition patterns. We found these forecasts were less accurate than our models.
Our models leverage historical malnutrition patterns, as well as clinical indicators and satellite-based indicators.
The forecasting performance of our models is also better than other similar data-based modelling efforts published by other researchers.
As resources for health and nutrition shrink, improved targeting to the areas of highest need is critical. Treating acute malnutrition can save a child’s life.
Prevention of malnutrition promotes children’s full psychological and physical development.
What needs to happen next
Making these data from diverse sources available through a dashboard could inform decision-making. Responders could get six months to intervene where they are most needed.
We have developed a prototype dashboard to create visualisations of what responders would be able to see based on our model’s sub-county-level forecasts. We are currently working with the Kenyan Ministry of Health and Amref Health Africa, a health development NGO, to ensure that the dashboard is available to local decision-makers and stakeholders. It is regularly updated with the most current data and new forecasts.
We are also working with our partners to refine the dashboard to meet the needs of the end users and promote its use in national decision-making on responses to acute malnutrition among children. We’re tracking the impacts of this work.
Throughout this process, it will be important to strengthen the capacity of our partners to manage, update and use the model and dashboard. This will promote local responsiveness, ownership and sustainability.
Scaling up
The Kenya Health Information System relies on the District Health Information System 2 (DHIS2). This is an open source software platform. It is currently used by more than 80 low- and middle-income countries. The satellite data that we used in our models are also available in all of these countries.
If we can secure additional funding, we plan to expand our work geographically and to other areas of health. We’ve also made our code publicly available, which allows anyone to use it and replicate our work in other countries where child malnutrition is a public health challenge.
Furthermore, our model proves that DHIS2 data, despite challenges with its completeness and quality, can be used in machine learning models to inform public health responses. This work could be adapted to address public health issues beyond malnutrition, like changes in patterns of infectious diseases due to climate change.
Laura Ferguson – Associate Professor, Population and Public Health Sciences, University of Southern California
Bistra Dilkina – Associate Professor of Computer Science, University of Southern California.
This work was a collaboration between the University of Southern California’s Institute on Inequalities in Global Health and Centre for Artificial Intelligence in Society, Microsoft, Amref Health Africa and the Kenyan Health Ministry.
Study details
Forecasting acute childhood malnutrition in Kenya using machine learning and diverse sets of indicators
Girmaw Abebe Tadesse, Laura Ferguson, Caleb Robinson, et al.
Published in PLOS One on 14 September 2025
Abstract
Objectives
Malnutrition is a leading cause of morbidity and mortality for children under-5 globally. Low- and middle-income countries, such as Kenya, bear the greatest burden of malnutrition. The Kenyan government has been collecting clinical indicators, including on malnutrition, using District Health Information Software-2 (DHIS2) for over a decade. We aim to address the existing gap in decision-makers’ ability to develop and utilise malnutrition forecasting capabilities for timely interventions. Specifically, our objectives include: develop a spatio-temporal machine learning model to forecast acute malnutrition among children in Kenya using DHIS2 data, enhance forecasting capability by integrating external complementary indicators, such as publicly available satellite imagery-driven signals, and forecast acute malnutrition at various stages and time horizons, including moderate, severe, and aggregated cases.
Methods
We propose a framework to forecast malnutrition risk for each sub-county in Kenya based on clinical indicators and remote sensory data. To achieve this, we first aggregate clinical indicators and remotely sensed satellite data, specifically gross primary productivity measurements, to the sub-county level. We then label the rate of children diagnosed with acute malnutrition at the sub-county level using the standard Integrated Food Security Phase Classification for Acute Malnutrition. We then apply and compare several methods for forecasting malnutrition risk in Kenya using data collected from January 2019 to February 2024. As a baseline, we used a Window Average model, which captures the current practice at the Kenyan Ministry of Health. We also trained machine learning models, such as Logistic Regression and Gradient Boosting, to forecast acute malnutrition risk based on observed indicators from prior months. Different metrics, mainly Area Under Receiver Operating Characteristic Curve (AUC), were used to evaluate the forecasting performance by comparing their forecast values to known values on a hold-out test set.
Results
We found that machine learning based models consistently outperform the Window Average baselines on forecasting sub-county malnutrition rates in Kenya. For example, the Gradient Boosting model achieves a mean AUC of 0.86 when forecasting with a 6-month time horizon, compared to an AUC of 0.73 achieved by the Window Average model. The Window Average method particularly fails to correctly forecast malnutrition in parts of West and Central Kenya where the acute malnutrition rate is variable over time and typically less than . We further found that machine learning models with satellite-based features alone also outperform Window Averaging baselines, while not needing clinical data at inference time. Finally, we found that recently observed outcomes and the remotely sensed data are key indicators. Our results demonstrate the ability of machine learning models to accurately forecast malnutrition in Kenya at a sub-county level from a variety of indicators.
Conclusions
To the best of the authors’ knowledge, this work is the first to use clinical indicators collected via DHIS2 to forecast acute malnutrition in childhood at the sub-county level in Kenya. This work represents a foundational step in developing a broader childhood malnutrition forecasting framework, capable of monitoring malnutrition trends and identifying impending malnutrition peaks across more than 80 low- and middle-income countries collecting similar DHIS2 datasets.
See more from MedicalBrief archives:
Call for action after 155 children die of malnutrition this year
Ethiopian healthcare buckles under malnutrition crisis
Treatment of malnourished children challenged: UK-MSF study
SA’s under-fives hungry, neglected and dying – Child Gauge 2024