Enhancing credit card fraud detection with a stacking-based hybrid machine learning approach.

Area of Science:

• Computer Science
• Data Science
• Financial Technology

Background:

• Cyber fraud complexity is increasing due to technological advancements.
• Conventional fraud detection methods struggle with evolving fraud tactics, leading to inefficiencies and false positives.
• The banking sector faces significant challenges in reliably identifying fraudulent credit card transactions.

Purpose of the Study:

• To develop an innovative stacking hybrid machine learning (ML) approach for enhanced credit card fraud detection.
• To overcome the limitations of conventional detection methods by integrating diverse ML models.
• To improve the accuracy and efficiency of identifying fraudulent financial transactions.

Main Methods:

• Implemented a stacking ensemble framework combining Decision Trees (DT), Random Forests (RF), Support Vector Machines (SVM), XGBoost, CatBoost, and Logistic Regression (LR).
• Employed sophisticated data preprocessing techniques including correlation-based feature selection and Principal Component Analysis (PCA) for efficiency.
• Utilized a meta-model to consolidate predictions from base models for superior performance.

Main Results:

• The stacking ensemble model achieved a high F1-score of 88.14% on a credit card transaction dataset.
• Demonstrated superior predictive performance compared to individual machine learning models.
• Effectively balanced precision and recall for robust fraud detection.

Conclusions:

• Ensemble methods, particularly stacking, are crucial for achieving strong and dependable cyber fraud detection.
• The proposed stacking hybrid ML approach significantly enhances the security of financial transactions.
• This method offers a promising solution to the escalating challenge of credit card fraud in the banking sector.