This study investigates the application of deep learning techniques and automated feature engineering in credit card fraud detection. The research employs a Convolutional Neural Network (CNN) model integrated with Deep Feature Synthesis (DFS) to enhance the accuracy and efficiency of fraud detection systems. A comprehensive dataset of credit card transactions is utilized, comprising 2,453,620 records with 1,432 fraudulent cases. The methodology involves extensive data preprocessing, including class imbalance handling and feature encoding. The DFS algorithm generates 118 new features from the original 25, capturing complex relationships within the data. The CNN model's performance is compared against traditional machine learning algorithms such as Logistic Regression, Random Forest, Support Vector Machine, and XGBoost. Results demonstrate that the CNN model with DFS-generated features significantly outperforms other approaches, achieving an accuracy of 91%, precision of 92%, recall of 90%, and an F1-score of 0.91. The study highlights the synergistic effect of combining deep learning architectures with advanced feature engineering techniques in addressing the challenges of credit card fraud detection. The findings contribute to developing more robust and adaptable fraud detection systems, potentially reducing financial losses and enhancing security in electronic transactions. Future research directions include exploring model interpretability, real-time application, and extension to multi-class fraud detection scenarios.