Low birth weight (LBW) is a critical indicator of neonatal health that strongly determines morbidity and long-term developmental issues. India bears a disproportionately high burden of LBW, especially among socio-economically disadvantaged populations. Early detection of LBW risk factors enables timely and effective interventions to minimize the burden. The study therefore seeks to introduce a novel combination of categorical data transformation and machine-learning (ML)-based predictive modeling on NFHS-5 data for newborns at risk of LBW in India. Based on available data from over 175,000 singleton births, exposure variables were grouped into four domains: socio-economic, prenatal, maternal morbidity, and maternal nutritional factors. CATPCA revealed moderate-to-high component loadings for maternal morbidity, prenatal care, and nutrition, while ANN modeling identified the same determinants as the strongest factors, with socio-economic factors and dietary irregularities being less influential. The ANN achieved 94.6% accuracy, representing strong predictive performance for identifying at-risk newborns.
The paper addresses the problem of estimating the population mean Y of the study variable Y in the presence of an auxiliary attribute f under stratified random sampling. Two general families of estimators for the population mean, using information on an auxiliary attribute, are proposed in a stratified sampling scheme. The expressions for the bias and mean squared error of the proposed families of estimators have been derived up to the first order of approximation. Optimal conditions are obtained under which the proposed families of estimators have minimum mean squared errors. The families of estimators of Shahzad et al. (2019) and Zaman and Kadilar (2020) are members of the proposed families. A numerical illustration is also provided to support the theoretical findings.
