Enhancing credit card fraud detection using traditional and deep learning models with class imbalance mitigation

Area of Science:

• Data Science
• Machine Learning
• Financial Technology

Background:

• Increasingly sophisticated fraudulent activities pose significant challenges to financial transaction security.
• Robust fraud detection is crucial for mitigating substantial financial losses.

Purpose of the Study:

• To evaluate and compare the effectiveness of various machine learning models for financial fraud detection.
• To address class imbalance and improve predictive accuracy in fraud detection systems.

Main Methods:

• Comparative analysis of logistic regression, decision tree, and random forest models.
• Development of a deep learning model with focal loss for enhanced detection.
• Application of Synthetic Minority Over-Sampling Technique (SMOTE) for class imbalance and hyperparameter tuning.

Main Results:

• Random forest model achieved 99.95% accuracy, 0.8256 F1 score, and 0.9759 ROC-AUC.
• Deep learning model demonstrated the highest precision, effective in minimizing false positives.
• Validation across both Kaggle credit card and PaySim synthetic mobile money datasets.

Conclusions:

• The random forest model offers excellent overall performance for fraud detection.
• Deep learning with focal loss shows promise for precise identification of fraudulent transactions.
• Combining data preprocessing, resampling, and model optimization yields robust fraud detection capabilities.