Fraud detection in e-commerce faces great challenges due to data imbalance, where legitimate transactions far outnumber fraudulent transactions. This research explores the use of Conditional Generative Adversarial Network (CGAN) to generate synthetic fraudulent transaction data to address the imbalance problem. By increasing the amount of data in the minority class, this research aims to improve the performance of two widely used machine learning algorithms, namely Random Forest and XGBoost. The dataset used of 23,634 transactions with 22,412 non-fraud transactions and 1,222 fraudulent transactions. Accuracy, precision, recall, and F1-score metrics were conducted to assess the performance of the model in detecting fraud on the imbalanced and augmented datasets. The results show that augmentation of data with CGAN significantly improves the performance of both models, especially in improving recall for fraudulent transactions. On the original unbalanced dataset, Random Forest and XGBoost showed low recall (12.81% and 13.08%), with accuracy of 95.35% and 95.32% respectively. However, after augmentation, recall improved to 95.15% for Random Forest and 95.22% for XGBoost, with F1-score of 97.47% and 97.42% respectively, and accuracy of 97.50% for Random Forest and 97.42% for XGBoost. XGBoost showed a slight advantage in precision and recall over Random Forest, especially on the augmented dataset. These findings confirm the effectiveness of CGAN as a data augmentation method in improving fraud detection performance and offer a robust solution to address data imbalance in the financial sector.