A Machine Learning Approach for Predicting Stunting in Under Five Children: The Case of Pakistan Demographic and Health Survey

Abstract Stunting in children less than five years of age is widely recognized as a major health problem in most of the developing countries of the world including Pakistan. It is considered as one of the contributing factors of death and multiple diseases. Pakistan has been reported to have one of the highest levels of prevalence of child malnutrition as compared to other developing countries as four out of ten children are stunted. Studies regarding prediction of nutrition status of children and identification of factors that lead to stunting have the potential to reveal great insights in the domain of healthcare informatics. This study identifies stunting in under-five children and also finds the association of demographics, socioeconomic and maternal characteristics that leads to stunting. The study is validated using nutrition-related attributes from Pakistan Demographic and Health Survey dataset. Machine learning based data driven model is trained to classify a child as normal or stunted. Results reveal that out of the selected attributes residence, wealth index, mother’s education, child’s age, and height are strong predictors of stunting. Results also show that the model can be used to predict various nutrition related problems in children using the survey dataset. The analysis of several evaluation metrics concludes that among four classifiers SVM outperforms with an average accuracy of 98.5%.