Fraud Detection in Mobile Payment Systems using an XGBoost-based Framework.

Area of Science:

• Financial technology
• Machine learning applications
• Cybersecurity

Background:

• Mobile payment systems are increasingly popular, leading to a rise in fraudulent activities.
• Existing supervised machine learning fraud detection methods struggle with insufficient labeled data and severe class imbalance in financial datasets.
• The financial consequences of fraud detection systems are critical for organizational decision-making.

Purpose of the Study:

• To propose an XGBoost-based fraud detection framework tailored for mobile transactions.
• To evaluate the framework's performance considering the financial impact of fraud detection.
• To compare the proposed methods against existing techniques for imbalanced data and outlier detection.

Main Methods:

• Development of a semi-supervised ensemble model integrating unsupervised outlier detection algorithms and an XGBoost classifier.
• Empirical validation on a large dataset comprising over 6 million mobile transactions.
• Comparative analysis with established machine learning methods for imbalanced data and outlier detection.

Main Results:

• The proposed semi-supervised ensemble model achieved superior performance based on standard classification metrics.
• Combining random under-sampling with XGBoost methods resulted in the highest cost savings.
• The study highlights the effectiveness of XGBoost in handling imbalanced financial fraud data.

Conclusions:

• The developed XGBoost-based framework offers an effective solution for mobile payment fraud detection, particularly with imbalanced datasets.
• Organizations can leverage these findings to make informed decisions on implementing cost-effective fraud detection systems.
• The research contributes to improving the accuracy and financial efficiency of fraud detection in mobile payment ecosystems.