Budi Prasetiyo, M. Faris Al Hakim, Fariska Ratna Fauziah
E-commerce provides great benefits to small businesses, individuals, and large corporations by allowing them to reach a wider audience to market their products and services, something that is difficult to achieve through traditional offline sales methods. However, alongside these benefits comes the challenge of an increased risk of fraud, which can undermine consumer trust and threaten the security of transactions. Therefore, fraud detection becomes an important component in ensuring the sustainability of the e-commerce ecosystem. This research develops a fraud detection model by combining K-means SMOTE oversampling technique and XGBoost algorithm optimized through two hyperparameter tuning methods, namely Random Search and Bayesian Optimization. The results show that the XGBoost model optimized with Bayesian Optimization produces the best performance, with an accuracy of 95.56 %. The K-means SMOTE technique effectively addresses data imbalance by creating representative synthetic samples. Meanwhile, Bayesian Optimization successfully identifies optimal parameters through iterative and precise search. These results demonstrate that the proposed approach not only effectively detects fraud but also maintains the stability of model performance. © 2025 IEEE.
Department of Computer Science, Universitas Negeri Semarang, Semarang, Indonesia
