Coupled modular simplicial graph neural network with snow ablation optimization for real-time fraud detection in payment systems
Open Access
- 16 February 2026
- journal article
- research article
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Abstract
The exponential growth of online transactions has increased the vulnerability of fraudulent activities in payment systems and necessitated sophisticated and smart detection techniques for real-time fraud protection. Conventional machine learning techniques tend to perform poorly with high-dimensional data, skewed classes, and evolving fraud patterns, which affect detection accuracy. In order to overcome these shortcomings, the study presents a new model, which is Coupled Modular Simplicial Graph Neural Network with Snow Ablation Optimization (CMSGNN-SAO) to effectively detect fraud in real-time. The process starts with pre-processing, where raw data from the Credit Card Fraud Detection (CCFD) dataset are processed using the Adaptive Morphological Wavelet Perona-Malik (AMWPM) filtering algorithm to remove noise, normalize features, and maintain data quality. Next, Feature Selection using Quokka Swarm Optimization (QukSO) is used to remove unnecessary features by keeping the most informative attributes and penalizing redundant or irrelevant ones. In the classification process, the Coupled Modular Simplicial Graph Neural Network (CMSGNN) is utilized, which inherits the advantages of Coupled Modular Neural Networks (CMNN) for modular learning and the Simplicial Graph Attention Network (SGAN) for efficient learning of higher-order topological relationships between transaction data. To enhance more accurately make predictions, the architecture adds Snow Ablation Optimization (SAO), which is the optimization of weights and reduction of misclassification error.The CMSGNN-SAO architecture yields enhanced flexibility, scalability, and reliability in detecting fraud versus non-fraud transactions. Experimental findings confirm its advantage in precision (99.1), recall (99.4), F1-score (99.2), accuracy (99.5), specificity (99.3), and ROC performance, making it a competent deep learning approach for real-time fraud detection (RTFD) in contemporary payment systems.Keywords
Funding Information
- Symbiosis International
This publication has 23 references indexed in Scilit:
- Embedding Accountability in the AI Lifecycle for Critical Finance Applications2025
- Hybrid CNN-LSTM With Attention Mechanism for Robust Credit Card Fraud DetectionIEEE Access, 2025
- Credit Card Fraud Detection Using Machine Learning Techniques: Dealing with Imbalanced Data using Over-Sampling and Under-Sampling Methods2024
- A Hybrid Deep Learning Ensemble Model for Credit Card Fraud DetectionIEEE Access, 2024
- Performance of Snow Ablation Optimization for Solving Optimum Allocation of Generator UnitsIEEE Access, 2024
- Online Payment Fraud Detection Model Using Machine Learning TechniquesIEEE Access, 2023
- An effective fraud detection using competitive swarm optimization based deep neural networkMeasurement: Sensors, 2023
- Credit Card Fraud Detection Using State-of-the-Art Machine Learning and Deep Learning AlgorithmsIEEE Access, 2022
- CMNN: Coupled Modular Neural NetworkIEEE Access, 2021
- Big-data driven approaches in materials science for real-time detection and prevention of fraudMaterials Today: Proceedings, 2021