Analysis of Data Balancing Techniques in Fraudulent Transactions Datasets

In the preprocessing stage that prepares data to be used in training a machine learning model, the available datasets are usually highly unbalanced. This paper analyzes and compares three methods for solving the issue of imbalanced datasets, using data from electronic payment transactions, where fraudulent operations represent the minority class. The Python implementation of the methods proves that the SMOTE with Tomek Links technique provides the best results, based on analyzing F1-scores, Precision-Recall and ROC curves.