Smart Detection of Credit Card Fraud Using Machine Learning

Abstract

The increase in digital financial transactions has dramatically increased credit card fraud, which now poses significant threats to consumers as well as financial institutions. This study provides a comparative study of a variety of machine learning models for credit card fraud detection using a real life, imbalanced dataset. We compared four machine learning models—Logistic Regression (LR), Random Forest (RF), XGBoost, and Artificial Neural Network (ANN)—in terms of their ability to detect fraudulent transactions accurately. We used the Synthetic Minority Over-sampling Technique (SMOTE) for preprocessing and balancing the datasets, but tested the machine learning models on the original imbalanced set to adequately represent real-world performance. The experimental results of this study indicate that the ANN and XGBoost models were the most accurate and had the highest recall and F1-score, with the ANN performing the best in most corrective metrics. The feature importance plot shows that there are some PCA variables, such as V14 and V17, that are relevant to fraudulent activity. The research provides evidence of the capabilities of ensemble and deep learning models in performing fraud detection tasks, especially after careful preprocessing and addressing problems related to the dataset imbalance. This study also provides a computationally sound and useful methodology that could be utilized in developing intelligent fraud detection systems in the financial sector.