Child Malnutrition, Neonatal and Infant Mortality: New Insights Learnt From Machine Learning


This paper investigates neonatal mortality and infant mortality in India using a large household survey data and by employing multiple parametric and non-parametric “Statistical Learning” techniques. The Statistical Learning methods address two common econometric challenges-predicting ‘rare-events’ and obtaining valid inference in the presence of many correlated covariates. Several Machine Learning (ML) methods along with traditional logistic regression are used to predict the incidences of neonatal and infant mortality, with Learning methods displaying substantially higher prediction accuracy than the conventional logistic model. Using the results this paper tries to identify a ‘high-risk mortality group’ for babies to help implement more targeted health policies. Additionally, a selection and shrinkage-based machine-learning technique (LASSO) is used to select from a large set of predictors for infant malnutrition measured by weight-for-length Z score. Inference is performed on the selected model using Statistical Learning based post-selection inference method. Based on these exercises we try to assess as well as recommend certain policy measures for combating child mortality and malnutrition.

Dweepobotee Brahma
Ph.D. Candidate in Applied Economics