The exponential growth in the number of online financial transactions has made the fight against fraud urgent. The purpose of this research is to determine how well ensemble machine learning models detect fraudulent actions using a synthetic transactional dataset. Our analysis and comparison focused on five models: Random Forest, Gradient Boosting, AdaBoost, Bagging, and Voting. The Bagging model outperformed the competition, displaying its resilience in the face of class imbalance, with a recall of 85%, a precision of 95%, and an accuracy of 99.98% in detecting fraud. Although it had a slightly lower recall (78% vs. 99.97%), Random Forest still did very well. The recall rates of Gradient Boosting (36% accuracy) and AdaBoost (30%) were poor, while they were accurate overall. Bagging is useful for unbalanced datasets, and these results show how important ensemble techniques are for detecting financial.