Evaluation performance recall and F2 score of credit card fraud detection unbalanced dataset using SMOTE oversampling technique
Budi Prasetiyo, Alamsyah Alamsyah, Much Aziz Muslim, Niswah Baroroh
Abstract
Abstract Unbalanced data becomes an interesting research and continues to be studied because of its uniqueness. Unbalanced data requires special treatment prior to making the data balance. In this paper, our study to investigate the performance of unbalanced dataset using diverse oversampling proportion. We use SMOTE to gerentae new syntethic data, then we classify using random forest algorithm. In our experiment we generate new sampling with start 20%, 40%, 60%, 80%, and 100% of majority class, so that the data balancing until 50%: 50%. Each new generated data, we train the data using classification technique. Then, evaluate each algorithm performance. We show that the highest F2 score i.e: 85.34 and 84.93. The new data generated is 60% of majority class, result F2 score 85.34, then the new data generated from 100% of majority class result F2 score 84.93.