Ensemble Model Approach For Imbalanced Class Handling on Dataset
Yoga Pristyanto, Anggit Ferdita Nugraha, Irfan Pratama, Akhmad Dahlan · 2020
In the field of machine learning, the distribution of classes in the dataset is important to produce a good model. The existence of class imbalances in the dataset is often ignored by researchers in the field of machine learning. This will certainly make the resulting model have less than maximum performance because theoretically, the single classifier has a weakness to the class imbalance conditions in datasets, this is because the majority of single classifier tends to work by recognizing patterns in the majority class in datasets that are not balanced so that the performance is not can be the maximum. Therefore, it is necessary to deal with these problems. In this study proposing an algorithmic level approach using Random Forest and Stacking to deal with these problems, the basic idea of using an algorithmic level approach is not to change the composition or pattern contained in the dataset itself. Based on tests using 5 datasets with different imbalance ratios, it shows that Random Forests or Bagging Tree and Stacking Naïve Bayes and Decision Tree C4.5 can produce better performance than single classifiers such as SVM, Naïve Bayes, and Decision Tree C4.5. So, the proposed method can be a solution in handling class imbalance in the dataset with different imbalance ratios.