<![CDATA[The use of electronic money (e-money) in Indonesia has grown rapidly, with transaction values reaching IDR 39.8 trillion in 2024. However, this growth is accompanied by an increase in fraud cases that harm consumers and service providers. This study aims to implement the Random Forest algorithm to detect fraud in e-money transactions. The research data uses the PaySim dataset, consisting of 1,048,575 transactions with 11 features and the isFraud label. The research method includes data preprocessing, handling imbalanced data using SMOTE (Synthetic Minority Oversampling Technique), Random Forest implementation with hyperparameter optimization using Grid Search, and model evaluation using precision, recall, and F1-score metrics. The results showed that the optimized Random Forest model achieved an accuracy of 99.92%, precision of 0.95, recall of 0.93, and F1-score of 0.94 on the test data. Feature importance analysis identified that the amount, oldbalanceDest, and newbalanceDest features are the most significant factors in detecting fraudulent transactions, with contributions of 35%, 28%, and 22%, respectively. The resulting model is able to detect 93% of fraudulent transactions with a prediction error rate of only 5%. Keywords: Fraud Detection; E-Money; Random Forest; SMOTE; PaySim]]>
Copyrights © 2026
