Transaction Fraud Detection Using SMOTE Oversampling

Credit card fraud detection remains a critical challenge in financial transactions, particularly due to the prevalence of unbalanced datasets and the limitations of traditional algorithms. This paper suggests a novel hybrid ensemble approach that reduces false negatives and improves fraud detection accuracy by merging the Random Forest (RF) and XGBoost (XGB) algorithms. Unbalanced data presents challenges for traditional algorithms such as Random Forest, Decision Trees, XGBoost, Logistic Regression, and Deep Neural Networks (DNN). Our hybrid technique, which is enhanced by under-sampling and the Synthetic Minority Over-sampling Technique (SMOTE), capitalizes on the benefits of RF and XGB while limiting their drawbacks. We prioritize recall over accuracy in order to reduce false negatives, which are important for detecting fraud. With an emphasis on recall, we evaluate the performance of Decision Trees, Random Forest, Logistic Regression, XGBoost, DNN, and our hybrid model. Utilizing techniques to address data imbalance, the evaluation makes use of an actual credit card dataset with different attributes. The results demonstrate the effectiveness of our hybrid ensemble approach in enhancing fraud detection accuracy and reducing false negatives, highlighting its potential for practical application in financial security systems.