KOMPARASI METODE SMOTE DAN ADASYN UNTUK PENANGANAN DATA TIDAK SEIMBANG MULTICLASS

Abstract

Data Mining is an activity that combines various branches of science into one, consisting of database systems, statistics, machine learning, and visualization, to analyze a large dataset in order to obtain useful data characteristics. To address the problem of imbalanced datasets, the distribution of non-uniform classes among classes is balanced by using a comparison of the SMOTE and ADASYN methods to ensure that the number is balanced between majority (negative) and minority (positive) classes. Based on the results of experiments conducted in this study, testing the SMOTE method with a classification method can handle the number of majority (negative) and minority (positive) classes in imbalanced data by producing MCC and Gmean values that achieve better predictive performance than using a classification method alone or using the ADASYN method. Furthermore, for the MultiClass dataset, the highest MCC and Gmean values were achieved using SMOTE + KNN with the highest MCC value of 0.64 and Gmean value of 0.74. This indicates that the handling process of imbalanced class distribution in the data preprocessing stage has an influence on the accuracy values of MCC and Gmean in the SMOTE + KNN method.