Improving Stunting Classification Performance Using Combination Smote Technique and Artificial Neural Network Algorithm

Child development is at the core of the nation's future. However, there are still serious problems that hinder optimal child growth, one of which is stunting. Stunting is a condition that has become a global concern in the context of public health and development. The use of deep learning algorithms has great potential to overcome the problem of stunting classification. The ratio of stunting handling is still a problem due to imbalance data. Classification algorithms such as ANN will experience a decrease in performance when faced with unbalanced classes, this makes it difficult to take action for early diagnosis of stunting. Synthetic Minority Oversampling Technique (SMOTE) is used to balance the failure data in diagnosis. The results showed that the combination of the SMOTE oversampling technique can improve the ability of the ANN algorithm model to accurately classify stunted or minority populations. The accuracy, precision, recall, and F1-Score values of this study are 0.90, 0.85, and 0.95, respectively. The results of (MCC) obtained a value of 0.73, and (G-Mean) of 0.86 before applying SMOTE and the results after applying SMOTE MCC of 0.84 and G-Mean of 0.92. This indicates that the minority group, namely stunted toddlers, can be predicted well. The implementation of the combination of SMOTE and ANN has proven successful in classifying imbalance stunting data, so this research can be used as a reference for future research to handle unbalanced data.