The widespread use of online transactions has revolutionized modern commerce but has also led to an increase in fraudulent activities, particularly in credit card transactions. To address these challenges, effective fraud detection systems are essential for safeguarding user data and minimizing financial losses. This study investigates the performance of multiple supervised machine learning algorithms, including Random Forest, XGBoost, LightGBM, CatBoost, and a novel Stacking Ensemble model, for detecting credit card fraud. Additionally, dimensionality reduction technique and a data balancing method are employed to enhance model efficacy. Comprehensive evaluations are conducted using metrics such as precision, accuracy, F1-score, and ROC-AUC. By identifying the most effective approach, this research contributes a robust and scalable fraud detection frame- work for real-world applications, advancing the reliability and security of financial systems.