Online Payment Fraud Detection Model Using Machine Learning Techniques
Модель обнаружения мошенничества при онлайн-платежах с использованием методов машинного обучения
2023-01-01
SCID: 54.1/3eegjz7b
Discuss with AI
Autoencoder-ResNet feature extractionJaya optimization algorithmOnline payment fraud detectionResNeXt-GRU modelSynthetic Minority Over-sampling Technique (SMOTE)
Figures from the paper
Abstract (AI)
In a world where wireless communications are critical for transferring massive quantities of data while protecting against interference, the growing possibility of financial fraud has become a significant concern. The ResNeXt-embedded Gated Recurrent Unit (GRU) model (RXT) is a unique approach precisely created for real-time financial transaction data processing. Motivated by the need to address the rising threat of financial fraud, which poses major risks to financial institutions and customers, our technique takes a systematic approach. We initiate the process with data input and preprocessing, addressing data imbalance through the Synthetic Minority Over-sampling Technique (SMOTE). Feature extraction uses an ensemble approach that combines autoencoders and ResNet (EARN) to reveal critical data patterns, while feature engineering further enhances the model’s discriminative capabilities. The core of our classification task lies in the RXT model, fine-tuned with hyperparameters using the Jaya optimization algorithm (RXT-J). Our model undergoes comprehensive evaluation on three authentic financial transaction datasets, consistently outperforming existing algorithms by a substantial margin of 10% to 18% across various evaluation metrics while maintaining impressive computational efficiency. This pioneering research represents a significant advancement in the ongoing battle against financial fraud, promising heightened security and optimized efficiency in financial transactions. In defense against wireless communication interference, our work aims to strengthen security, data availability, reliability, and stability against cyber warfare attacks within the financial industry.
Key Findings
1
Across three authentic financial transaction datasets, RXT-J outperformed existing algorithms by 10% to 18% across multiple evaluation metrics.
2
SMOTE addresses class imbalance, while the EARN ensemble combines autoencoders and ResNet for discriminative feature extraction.
3
The RXT classifier is hyperparameter-optimized using the Jaya algorithm, forming the RXT-J fraud detection model.
4
The model maintained computational efficiency while improving fraud detection performance, supporting secure and reliable financial transaction processing.
5
The proposed RXT model combines ResNeXt embeddings with gated recurrent units for real-time financial transaction fraud detection.
Research Object
Online financial transactions and their transaction data
Research Subject
real-time fraud detection performance, including discriminative accuracy and computational efficiency
Publication Details
Publication Date
2023-01-01
Journal
Publisher
ISSN
Cited by
147
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest