The rise of online payments has led to increased online fraud, with traditional rule-based detection systems struggling to keep up. Machine learning (ML) has emerged as a more effective solution, but existing research often focuses on a limited set of models or fails to properly address class imbalance in fraud detection datasets. This study aims to fill these gaps by comparing various ML classifiers on both imbalanced and balanced datasets. Additionally, it explores the use of ensemble methods to improve fraud detection performance, particularly in identifying the minority class (fraud), and introducing our DROE model, providing a more comprehensive approach to this challenging problem.