Identifying Determining Factors of Low Birth Weight in India using Categorical Principal Component and Artificial Neural Network Model: An Insight from NFHS-5

Alokananda Ghosh, Tehatta Sadananda Mahavidyalaya

Background and objectives: Low Birth Weight (LBW) is one of the key indicators of infant morbidity, and mortality, especially in low-middle-income countries. Therefore, early recognition of LBW in infants is necessary to avoid the risk. The objective here is to develop a best predictive model, identifying the determining factors behind the occurrence of LBW in India. Methods: The study sample included women of reproductive age group (15-49 years) delivering singleton live-born baby (N=26366) in her most recent birth having LBW (<2500 grams) using NFHS-5 dataset (2019–21), a national level survey representative of India. The dimensions of the indicators have been reduced by using categorical principal component analysis (CATPCA). Four sets of putative predictive factors covering socio-economic, prenatal, morbidity and nutritional factors have been included to develop four 2-component CATPCA models. The varimax rotation method adopted here and the factors of dimension 1 and 2 with loading value = ±0.5 of each CATPCA model have been considered as strong determinants of the final Artificial Neural Network (ANN) predictive model. Results and conclusion: The main determining factors are- edema and convulsion without fever (model 3, dimension 2), marital age, under age first birth and current age of respondent (model 2, dimension 2).

Keywords: Health and Morbidity, Children, Adolescents, and Youth, Multi-level modeling , Civil Registration and Vital Statistics

See extended abstract.