A HYBRID MULTINOMIAL LOGISTIC REGRESSION-NEURAL NETWORK APPROACH TO MODELLING NUTRITIONAL STATUS OF NIGERIAN WOMEN OF REPRODUCTIVE AGE

Abstract

This study focuses on modelling the nutritional status of Nigerian women of reproductive age using
a hybrid approach combining multinomial logistic regression model with a neural network
component. A secondary data extracted from Multiple Indicators Cluster Survey (MICS 6: 2021-
2022) report was the source of data for the study while data analysis was carried out with the aid
of STATA Software version 16.0. Study results show that all coefficients of predictor variables
greater than 3.84 of chi-square distribution are statistically significant in the modified model with
respect to nutritional status of WRA. The study also revealed that all predictor variables with p −
value of less than 0.05 significance level are significantly associated with outcome variables
(underweight, overweight and obesity). By integrating multinomial logistic regression for
interpretability with neural network components for enhanced predictive accuracy, this study
provides a robust modelling framework. It was recommended that establishment of a

comprehensive database on women’s health and nutrition in Nigeria, incorporating socio-
demographic, economic and health-related factors after proper data collection is ensured.

Keywords: Nutritional Status, Women of Reproductive Age, Multinomial Logistic Regression,
Neural Network, Nigeria and Maternal Health.