Comparative Analysis of Machine Learning Models for Credit Card Fraud Detection in Imbalanced Datasets

Abstract

This study presents a comprehensive evaluation of various machine learning models for detecting credit card fraud, emphasizing their performance in handling highly imbalanced datasets. We focused on three models: Logistic Regression, Random Forest, and Multilayer Perceptron (MLP), using a dataset comprising 555,719 transactions, each annotated with 22 attributes. Logistic Regression served as a baseline, Random Forest was evaluated for its high accuracy and low dependency on hyperparameter tuning, and MLP was tested for its capability to identify non-linear patterns. The models were assessed using ROC AUC, Matthews Correlation Coefficient (MCC), and precision-recall curves to determine their effectiveness in distinguishing fraudulent transactions. Results indicated that the Random Forest model outperformed others with a ROC AUC of 0.9868 and an MCC of 0.6638, showing substantial superiority in managing class imbalances and complex data interactions. Logistic Regression, although useful as a benchmark, exhibited limitations with a high number of false positives. MLP showed potential but was prone to a significant false positive rate, suggesting a need for further model refinement. The findings highlight the importance of choosing appropriate models and feature engineering techniques in fraud detection systems and suggest avenues for future research in real-time model deployment and advanced algorithmic strategies