ANALYSIS OF MALNUTRITION AMONG NIGERIAN WOMEN OF REPRODUCTIVE AGE USING MULTINOMIAL LOGISTIC REGRESSION AND STATISTICAL MACHINE-LEARNING NEURAL NETWORK APPROACH

Abstract

Malnutrition among women of reproductive age (15 – 49 years) in Nigeria manifests heterogeneously as undernutrition, overweight and obesity with profound public-health implications. This study analyzed nationally survey data from the 2021 – 2022 Multiple Indicator Cluster Survey 6 (MICS 6) for Nigeria using multinomial logistic regression (MLR) and statistical machine-learning neural network approach. The results obtained from the modified model indicated that Apparent Prediction Error (APE) of 0.873 indicates that the modified model in this study correctly predicted 87.3% of the outcomes in the training datasets while Expected Prediction Error (EPE) provides a more realistic estimate of the model’s performance on large datasets, such that the Expected Prediction Error (EPE) of 0.878 obtained indicates that the modified model is expected to correctly predict approximately 87.8% of the outcomes in new, unseen data. This showed that the modified model has good generalizability and machine-learning capability with very low Standard Error of 0.003 which implies that variability of the estimate is low, an indication of better performance of the modified model employed in this study for prediction over the traditional multinomial logistic regression model. This modified model also prevented overfitting of the training data as data partioning to prevent overfitting in this study was 70:30, which implies that 70% of datasets for training while 30% for testing. Regular update and refinement of the improved model with new data and emerging factors to maintain its predictive accuracy relevance and integration of nutritional education with counseling initiatives in existing maternal and child health programs can empower women to make informed choices and prevent malnutrition-related complications were recommended.