Adaptive Fraud Detection: A Machine Learning Framework Combining Supervised and Unsupervised Learning Techniques

The widespread use of credit cards in financial transactions has significantly increased the risk of fraudulent activities. Detecting fraud in real time is a critical challenge due to the highly imbalanced nature of transaction datasets and the continuous adaptation of fraud strategies. This paper presents a comprehensive study of machine learning techniques for credit card fraud detection. The methodology includes dataset preprocessing, feature engineering, handling of class imbalance, and the application of both supervised and unsupervised learning algorithms. Models including Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, and Autoencoders were evaluated using performance measures such as Precision, Recall, F1-Score, and PR-AUC. Results indicate that Gradient Boosting achieves the most effective balance between fraud detection and false alarm reduction, while Autoencoders are effective in identifying emerging fraud patterns. The study emphasizes the importance of combining supervised and unsupervised methods for robust fraud detection and concludes with recommendations for future enhancements in real-world systems.