COMPARATIVE ANALYSIS OF MULTINOMIAL LOGISTIC REGRESSION AND NEURAL NETWORKS FOR MALNUTRITION PREDICTION AMONG NIGERIAN WOMEN OF REPRODUCTIVE AGE

Abstract

In Nigeria, malnutrition affects women between the ages of 15 and 49 in a variety of ways, including undernutrition, overweight, and obesity, all of which have significant effects on public health.  This study uses statistical machine-learning neural networks and multinomial logistic regression (MLR) to compare the nutritional status of Nigerian women of reproductive age. This study examined national survey data from Nigeria's Multiple Indicator Cluster Survey 6 (MICS 6) conducted in 2021–2022. In comparison to existing multinomial logistic regression with Apparent Prediction Error of 0.873 (87.3%) and Expected Prediction Error of 0.878 (87.8%) with Standard Error of 0.041, the results obtained from the formulated model showed that Apparent Prediction Error (APE) of 0.756 indicates that the formulated model in this study correctly predicted 75.6% of the outcomes while Expected Prediction Error (EPE) of 0.759 (75.9%) provides a more realistic estimate of the model's performance on large datasets. This demonstrated that the developed model performs better for prediction than the conventional multinomial logistic regression model and has good generalizability and machine-learning capability with a very low Standard Error of 0.003, which suggests that the variability of the estimate is low. In order to increase policy relevance, it was recommended that UNICEF and the National Bureau of Statistics encourage the use of hybrid statistical-machine learning models in routine analysis of MICS data.