PREDICTING NUTRITIONAL STATUS IN NIGERIAN WOMEN OF REPRODUCTIVE AGE: A HYBRID MULTINOMIAL LOGISTIC REGRESSION MODEL AND NEURAL NETWORK APPROACH

Abstract

This study developed a hybrid predictive model combining Multinomial Logistic Regression (MLR) and Neural Network (NN) Approach to assess and predict nutritional status among Nigerian WRA. Utilizing data from Multiple Indicator Cluster Survey 6 (MICS 6), the hybrid model evaluated the predictor variables of underweight, overweight and obesity using normal weight as the baseline category. Key determinants (predictor variables) include age, women education, access to mass media, availability of sufficient water when needed, iodized salt consumption, geo-political zone and ethnicity. Model evaluation performance indicated robust predictive accuracy of the hybrid model in this study which correctly predicted (APE) of the outcomes in the training datasets while the Expected Prediction Error (EPE) provides a more realistic estimate of the model’s performance on large datasets, as opposed to Apparent Prediction Error (APE) which is based on the training data. It was revealed in this study that the hybrid model Expected Prediction Error (EPE) is expected to correctly predicted approximately  of the outcomes in new, unseen data with very low Standard Error of  which implies that variability of the estimate is low which is an indication of better performance of the hybrid model developed in this study for prediction over the traditional multinomial logistic regression model. The study provides insights for targeted nutritional interventions and policy formulation to improve health outcomes for WRA in Nigeria. It is recommended that collaboration among government agencies, international organizations and local stakeholders is very essential to ensure effective implementation of nutrition policies and programmes.