• The main aim is to develop a spatio-temporal machine learning model for predicting acute malnutrition in children, using DHIS2 data.

• Main objectives of this study are improving the prediction rates through the collective use of all indicators, such as signals derived from satellite images, and forecasting acute malnutrition at various severity levels across different time periods of 1, 3, and 6 months.

• The study used routinely collected health data from DHIS2 alongwith remotely sensed satellite data, mainly Gross Primary Productivity measurements, of 5 years.

• Window average model has been used as a baseline and it is compared with the new machine learning model. The performance was evaluated using Area Under Receiver Operating Characteristic Curve (AUC).

• The machine learning models with AUC of 0.86 for 6 months period, performed much superior than the Window Average baseline of AUC 0.73 in forecasting sub-county malnutrition rates.

• The machine learning models displayed high accuracy of AUC >0.9 in predicting extreme acute malnutrition risk across all assessed time periods.

Click here to read more.