Multiclass Imbalanced Handling using ADASYN Oversampling and Stacking Algorithm

Abstract

Class imbalance conditions in datasets are common in real-world problems. Class imbalance is a condition where the number of classes in the dataset used in the classification process has a significant difference in number. In theory, most single classifiers have a weakness against class imbalance conditions in datasets, especially those with multiclass types, so their performance cannot be maximized. This study proposes two approaches to overcome the problem of multiclass imbalanced, namely the use of ADASYN (Adaptive Synthetic) Sampling and the Stacking Algorithm. As confirmed by testing on five multiclass datasets, the proposed method outperforms other methods in terms of accuracy values, sensitivity, specificity, and geometric mean values. As a result, the method proposed in this study can solve class imbalance problems in multiclass-type datasets. However, this study has limitations. Namely, the dataset used is a multiclass category with a maximum number of six classes. For this reason, further research will suggest testing using imbalanced class datasets in the category of multiclass datasets with more than six classes.