Comparative Analysis of Machine Learning Algorithms for Detecting Fraudulent Transactions in Highly Imbalanced Credit Card Data
Prachi Gupta, Shatakshi Shukla, Vaishali Kikan, Ashwni Kumar · 2024
This research investigates the effectiveness of various machine learning algorithms for detecting fraudulent transactions within a highly imbalanced dataset comprising credit card transactions from European cardholders in September 2013. The study employs four anomaly detection techniques: Isolation Forest, Local Outlier Factor (LOF),Logistic Regression and OneClass SVM, along with advanced ensemble methods such as XGBoost, to address the challenges posed by the dataset's skewness, where fraud cases constitute only 0.172% of transactions. Initial results indicate that the Isolation Forest algorithm outperforms others, demonstrating a detection accuracy of 99.38%, compared to 99.26% for LOF and 54.82% for SVM, highlighting its efficiency in isolating fraudulent transactions with fewer conditions. The use of XGBoost further complements this study by providing robust classification capabilities, particularly when enhanced by SMOTE for addressing class imbalance. These findings suggest that while traditional models offer considerable insights, integrating advanced machine learning techniques can significantly improve the detection of fraudulent activities in credit card transactions. This study underscores the importance of selecting appropriate models based on dataset characteristics and the potential of ensemble learning in improving prediction accuracy.