Classification of Imbalanced Data Using Random Forest Algorithm with SMOTE and SMOTE-ENN (Case Study on Stunting Data

The random forest algorithm is one of the widely used machine learning classification methods because it has the advantage of reducing the risk of overfitting while improving general prediction performance. However, for data with unbalanced classes, this algorithm lacks to achieve its best performance, particularly in predicting data in the minority class. As a result, this article proposes two resampling approaches to balance the data: the Synthetic Minority Oversampling Technique (SMOTE) and the Synthetic Minority Oversampling Technique with Edited Nearest Neighbors (SMOTE-ENN). For the data classification technique, the random forest algorithm is applied to the original data, then to the resampling results using both SMOTE as well as SMOTE-ENN. The case study was applied to stunting data consisting of 421 cases in the majority class and 79 in the minority class. An accuracy of 89% was obtained on the original data, 90% on the resampled data with SMOTE-ENN, and 91% on the resampled data with SMOTE.  The best accuracy was obtained using resampling technique with SMOTE, however it was not particularly significant.