Anomaly Detection Classifiers for Detecting Credit Card Fraudulent Transactions

The internet and e-commerce have grown quickly, which has increased credit card use but also, regrettably, increased credit card fraud. To address this, Anomaly Detection has emerged as a crucial method for identifying unusual events and data in datasets. It uses advanced algorithms to detect deviations from normal patterns, helping authorities proactively combat fraudulent activities. While digital advancements offer convenience, they also expose vulnerabilities. Anomaly Detection offers a modern defense, safeguarding financial systems by early spotting of anomalies. In this study, we employed two algorithms - the Isolation Forest (IF) and the Local Outlier Factor (LOF) for identification of anomalies. To improve the performance of these models, we also employed a variety of resampling strategies. Specifically, we used techniques like Random Undersampling, AllKNN, Synthetic Minority Oversampling Technique (SMOTE), and Synthetic Minority Oversampling Technique - Edited Nearest Neighbor (SMOTE-ENN) to balance the European Credit Card Fraudulent transactions dataset and the German Credit Card fraud dataset. Out of the different configurations, the Isolation Forest classifier demonstrated the highest accuracy, reaching 99.81%, when applied to the initially imbalanced European credit card fraudulence dataset. On the other hand, the German credit card dataset achieved a remarkable accuracy of 70.60% through the implementation of the LOF classifier, coupled with the Random Undersampling technique to address its imbalanced nature.