Credit Card Fraud Detection Using Machine Learning and Deep Learning Techniques

The rapid advancement of digital technologies and the widespread adoption of online payment systems have significantly transformed the financial sector. Credit cards have become one of the most widely used modes of transaction due to their convenience and speed. However, this growth has also led to a rise in fraudulent activities, resulting in substantial financial losses for both customers and financial institutions. Detecting fraudulent transactions is challenging due to the highly imbalanced nature of transaction data and the continuously evolving strategies of fraudsters. Therefore, developing an efficient and reliable fraud detection system is essential. This research focuses on designing an intelligent credit card fraud detection system using machine learning and deep learning techniques. The study utilizes a structured dataset containing anonymized numerical features along with a target variable indicating whether a transaction is fraudulent or legitimate. Data preprocessing techniques such as feature scaling and exploratory data analysis are performed to enhance data quality and understand feature distributions. A major challenge addressed in this study is class imbalance, where fraudulent transactions represent only a small portion of the dataset. To overcome this issue, the Synthetic Minority Over-sampling Technique (SMOTE) is applied to balance the dataset, enabling models to learn effectively from both classes. Several classification models, including Logistic Regression, K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Random Forest, and Convolutional Neural Network (CNN), are implemented and evaluated. The experimental results indicate that Random Forest achieved the highest accuracy of 99.68%, followed by Support Vector Machine with 99.06% accuracy, KNN with 96.25% accuracy, and Logistic Regression with competitive performance, while the CNN model achieved an accuracy of 20.63%, showing poor performance on structured tabular data. To ensure robustness and reliability, K-Fold Cross-Validation (k=5) is applied. The cross-validation results show consistent performance, with Random Forest achieving an average validation accuracy of approximately 99.5%, SVM around 98.9%, KNN around 95–96%, and Logistic Regression around 94–95%, confirming strong generalization capability across different data splits. Model performance is further evaluated using accuracy, precision, recall, and F1-score, along with confusion matrix analysis. To enhance interpretability, SHAP (SHapley Additive exPlanations) is employed, which reveals that features such as Time, V12, V14, and V10 significantly influence fraud detection decisions. The findings emphasize that ensemble models like Random Forest and SVM are highly effective for fraud detection, whereas deep learning models like CNN are less suitable for structured datasets. In conclusion, this study presents an effective, reliable, and interpretable framework for credit card fraud detection, contributing to improved financial security.