What question did this study set out to answer?

To explore the relationship between child undernutrition and human and sustainable development indices in Pakistan over 22 years.

March 10, 2026

Deep Neural Network Modeling of Sustainable Development and Human Development Indices on Child Undernutrition in Pakistan

Key Points

To explore the relationship between child undernutrition and human and sustainable development indices in Pakistan over 22 years.
Utilized deep learning frameworks including a DNN-MRM and a Bayesian neural network (BNN).
Applied the Autoregressive Distributed Lag (ARDL) bounds testing method for analysis.
Incorporated causal discovery with PC algorithm to derive structural priors.
Analyzed data from 2000 to 2022 to reflect changes before and after the Sustainable Development Goals.
Child undernutrition is significantly negative correlated with both HDI and SDI.
Higher HDI and SDI levels are linked to reduced child malnourishment.
DNN-MRM model achieved 95% accuracy compared to 82% accuracy of MRM model.
Bayesian network topology shows that child undernutrition reduction is associated with sustainable development, income, and health spending.

Abstract

ABSTRACT Child malnutrition is a profound public health problem in Pakistan and is pegged with the country's persistent struggles in human as well as sustainable development. The paper investigates the effect of Pakistan's pathways on the Human Development Index (HDI) and Sustainable Development Index (SDI), with respect to child under‐nutrition over a 22‐year period (2000–2022). We selected this period of time because it is meaningful, including the full period of the Millennium Development Goal and post‐Sustainable Development Goal agenda. In order to quantify this complicated relationship, we utilized a state‐of‐the‐art computational modelling method based on deep learning frameworks. Our main deep learning analytical approaches contained a multivariate regression model constructed (DNN‐MRM) built in a deep neural network architecture along with a Bayesian neural network (BNN) that had structural priors extracted through causal discovery using the PC algorithm. We also employed a time‐series econometric‐based Autoregressive Distributed Lag (ARDL) bounds testing approach as robust, securing the analytic rigor. The results suggest that undernutrition is significantly and negatively associated with HDI as well as SDI, which means high levels of both indices are related to declines in child malnourishment. Additionally, the structural parameters from Bayesian network topology demonstrate that child undernutrition reduction is conditionally associated with sustainable development, income, and health spending, especially given the percentage effect of human development. In DNN‐MRM model, ReLU activation function yielded optimal outcome, with the deep learning‐based BNN model significantly outperforming the MRM (95% vs. 82% accuracy). The results are therefore original and yield useful insights to the Pakistani literature. The present study suggests a multisectoral policy drive, but with an overriding emphasis on synergistic increase in SDI and HDI, especially concerning financial and health variables, for sustained reduction of child undernutrition.

Mark Helpful

Bookmark

Relay