Advanced fraud detection in financial systems: a comparative study of machine learning models on imbalanced data.

Area of Science:

• Machine Learning
• Data Science
• Financial Technology

Background:

• Credit card fraud poses a significant financial risk.
• Class imbalance is a common challenge in fraud detection datasets.
• Evaluating diverse machine learning models is crucial for effective fraud prevention.

Purpose of the Study:

• To comparatively evaluate multiple machine learning models for credit card fraud detection.
• To assess model performance using comprehensive metrics tailored for imbalanced datasets.
• To identify optimal models for distinguishing fraudulent from legitimate transactions.

Main Methods:

• Data preprocessing, feature scaling, and undersampling were applied to address class imbalance.
• Models evaluated include Logistic Regression, Support Vector Machine (SVM), Random Forest, XGBoost, Multi-Layer Perceptron (MLP), and TensorFlow/Keras Neural Network.
• Performance metrics included accuracy, precision, recall, F1 score, and AUC-ROC.

Main Results:

• XGBoost achieved the highest accuracy (96.6%) and a perfect precision score.
• MLP and TensorFlow/Keras neural networks demonstrated superior AUC scores (0.982 and 0.977, respectively).
• SVM, Logistic Regression, and Random Forest also exhibited strong performance with high precision and recall.

Conclusions:

• Ensemble methods (XGBoost) and neural networks are highly effective for credit card fraud detection in imbalanced datasets.
• The choice of evaluation metrics is critical and should align with specific application priorities, balancing false positives and negatives.
• This research provides insights into selecting robust machine learning models for financial fraud detection.