A novel method for detecting credit card fraud problems.

Area of Science:

• Machine Learning
• Data Science
• Cybersecurity

Background:

• Credit card fraud poses a significant financial threat, costing billions annually.
• Detecting fraud is challenging due to highly imbalanced datasets where legitimate transactions vastly outnumber fraudulent ones.
• Existing oversampling techniques often produce unrealistic or overgeneralized samples, hindering fraud detection efficacy.

Purpose of the Study:

• To propose a novel hybrid method, Autoencoder with Probabilistic XGBoost based on SMOTE and CGAN (AE-XGB-SMOTE-CGAN), for enhanced credit card fraud detection.
• To address the class imbalance problem in credit card fraud datasets using a synergistic two-phase oversampling approach.
• To improve the accuracy and reliability of fraud detection systems compared to existing machine learning algorithms.

Main Methods:

• Utilized an Autoencoder (AE) for feature representation learning to extract relevant patterns from the imbalanced dataset.
• Implemented a hybrid oversampling strategy combining Synthetic Minority Over-Sampling Technique (SMOTE) and Conditional Generative Adversarial Network (CGAN).
• Employed XGBoost classifier with a probabilistic threshold for the final classification of fraudulent transactions.

Main Results:

• The AE-XGB-SMOTE-CGAN algorithm demonstrated a 2% improvement in accuracy (ACC) compared to KNN and LightGBM.
• Achieved a 30% higher Matthew's Correlation Coefficient (MCC) compared to KNN at a 0.35 threshold.
• The method showed enhanced true positive and true negative rates, indicating superior performance in identifying both fraudulent and legitimate transactions.

Conclusions:

• AE-XGB-SMOTE-CGAN is a promising and effective method for detecting credit card fraud, outperforming conventional algorithms.
• The hybrid SMOTE-CGAN approach successfully generates realistic synthetic data, mitigating the class imbalance issue.
• The proposed method offers a robust solution for improving the accuracy and reliability of financial fraud detection systems.