Data mining based prediction of malnutrition in Afghan children

Proper nutrition is an essential component for the survival, growth, and development of children in society.  Malnutrition is a global problem in today’s life. The primary target of this phenomenon is children under five years of age and mostly for developing countries. In this study, a data mining approach is proposed to predict the malnutrition status of children under five years of age in Afghanistan. Random Forest, PART rule induction, and Naïve Bayes classifiers were applied to Afghanistan Nutrition SMART Survey data. The results were compared with Logistic Regression statistical method. Random Forest and PART rule induction classifiers performed well with the highest accuracy for stunted, underweight, wasted, and nutritional oedema statuses with and without transformed attributes. This study defines how data mining classification techniques can classify malnutrition status for children under five years of age. Generally, our method was the one that obtained the most robust results to predict the malnutrition status based on clinical sign and anthropometric parameters of preschool-age Afghan children.