Comparison of Elliptic Envelope Method and Isolation Forest Method on Imbalance Dataset

Abstract

The problem of unbalanced data is important in the field of Data Mining. Dataset with unbalanced classes is a dataset whose frequency of occurrence of certain classes is very much different from other classes. This imbalance problem will bias the classifier's performance. Many researchers have examined both the development of algorithms and modifications to the preprocessing stage to overcome this problem. This study discusses the comparison of One Class Classification algorithms, namely Elliptic Envelope and Isolation Forest on unbalanced data. From this study, the Elliptic Envelope Method showed better results compared to the Isolation Forest method with 80.28% recall testing and 80.28% precision while Isolation Forest showed 46.95% recall results and 46.95% precision.