Online Payment Fraud Detection Using the XGBoost Algorithm in MSMEs

Abstract

Abstract.  Rapid growth of online payment transactions among Micro, Small, and Medium Enterprises (MSMEs) in Indonesia has been followed by rising fraud cases that threaten business sustainability. Fraud in digital payments can cause major financial losses and weaken consumer trust in online payment systems. Therefore, MSMEs need an accurate and fast fraud detection system to identify suspicious transactions. This study develops a fraud detection model using the XGBoost (eXtreme Gradient Boosting) algorithm to detect fraudulent transactions in real time within MSMEs payment systems. The model shows high accuracy, precision, recall, and F1-score, outperforming baseline models such as Random Forest and Logistic Regression. The dataset used is the “Fraud Detection” dataset from Kaggle. The research process includes data preprocessing using SMOTE (Synthetic Minority Over-Sampling Technique) to address class imbalance, feature engineering to generate additional attributes such as transaction frequency, and data scaling with RobustScaler to handle outliers. Results show that the XGBoost model performs very strongly in detecting fraudulent transactions. The model also produces predictions quickly, making it suitable for real-time fraud detection in MSMEs online payment platforms. By identifying suspicious transaction patterns early, the system helps MSMEs reduce potential financial losses and maintain trust in digital payment services for users overall.