High-Resolution Prediction of Acute Malnutrition Prevalence Across Sub-Saharan Africa

Percentages of acute malnutrition continue to be unsettlingly high in developing countries, while coverage of treatment remains unsatisfactory. Here I develop software to predict acute malnutrition prevalence at a high resolution across sub-Saharan Africa. First I assembled a training dataset for machine learning of malnutrition prevalence. The 33 training variables may be split into five categories: development, economics, political situations, climate, and crop health. I then trained a random forest regressor on these features to predict acute malnutrition prevalence at a 5km resolution across sub-Saharan Africa. A validation with a train-test split showed that the algorithm could predict malnutrition prevalence with an average difference of 0.86\% prevalence, corresponding to an error of 10.7\%. The training features with the highest predictive power are, in order: female education, precipitation, forest cover, school attendance, and political stability. Next, acute malnutrition prevalence was forecasted to 2021. These predictions suggest that the prevalence of acute malnutrition will continue to decrease in coming years, but the number of acute malnutrition cases will actually increase due to population growth. This work highlights the importance of female education as well as the need for more cost-effective nutrition interventions as the number of acute malnutrition cases begins to rise.