Resampling and Ensemble Method for Enhance Breast Cancer Classification Performance in Imbalance Dataset
Anik Andriani, Anastasia Meyliana, Indriyanti, Vadlya Maarif, Akhmad Syukron · 2024
Imbalance dataset cases are obstacles that can affect the classification performance. Classification models on imbalanced datasets tend to predict the majority class, which can cause a bias. In addition, the classification accuracy can be misleading. The Breast Cancer Wisconsin dataset is an imbalanced dataset in which the composition of the two classes is unbalanced. This study applied several methods for handling imbalanced datasets, including resampling techniques and ensemble methods. The results show that classification on datasets that apply resampling with the Undersampling and Oversampling methods has increased the performance of the Decision Tree method. The ensemble method with the balanced random forest classifier provides the highest accuracy value compared to the classification results on datasets without handling imbalanced datasets or on classification with resampling techniques. Although the accuracy value of the classification with the application of the resampling technique is still below the results of the balance-random forest, both resampling techniques provide very good time execution.