Penerapan Metode Supervised Learning dan Teknik Resampling untuk Prediksi Penipuan Transaksi Keuangan

Financial transaction fraud can result in devastating consequences for the stability of companies, as well as huge losses for shareholders, the industry, and even the market as a whole. As fraud in financial transactions increases, there is a need for effective methods to accurately detect and prevent fraudulent activities. This study aims to compare the performance of five machine learning models, namely Random Forest, K-Nearest Neighbors (KNN), Decision Tree, XGBoost, and Extra Trees, in detecting financial transaction fraud using an imbalanced dataset. To overcome the data imbalance problem, three resampling techniques are applied, namely Synthetic Minority Oversampling Technique (SMOTE), Adaptive Synthetic Sampling (ADASYN), and Undersampling. Experiments were conducted with two training and test data sharing ratios, namely 70:30 and 80:20. The evaluation results showed that the XGBoost model was the most consistent, with the highest ROC AUC value of 99%, especially after the application of resampling techniques. The 80:20 data ratio resulted in a more balanced distribution and better model performance in detecting the minority class, particularly after resampling. This study concludes that the XGBoost model with resampling techniques is highly effective in addressing data imbalance.