ENSEMBLE BAGGING WITH ORDINAL LOGISTIC REGRESSION TO CLASSIFY TODDLER NUTRITIONAL STATUS

Abstract

One problem in classifying stunting data is that the data used does not have a balanced proportion. This study aims to apply the logistic regression classification method with ordinal scale response variables to overcome class imbalance through the ensemble bagging approach. The data used is secondary data in the form of final research reports that have been tested for validity and reliability. The predictor variables used are economic conditions, health services and the environment with categorical response variables, namely the nutritional status of toddlers in the categories of stunting, normal and high. The methods used are ordinal logistic regression and ensemble bagging on ordinal logistic regression with bootstraps of 100, 500, and 1000. The variables that influence the nutritional status of toddlers are Economic Conditions, Health Services, and the Environment. The results of the study showed that the accuracy, sensitivity, specificity, and F1-Score for ordinal logistic regression were smaller than ensemble bagging in ordinal logistic regression. The best classification method obtained was bagging logistic regression with a bootstrap number of 500 and obtained an accuracy value of 85%, sensitivity of 87.2%, specificity of 72.6%, and F1-Score of 79.3%.