Scholarlink

<p>Financial fraud in business environments has evolved in response to the growth of digital transactions, posing a more significant threat to data security and resulting in substantial economic losses. The primary issue with current fraud detection systems is that they are either overly rule-based or too reliant on classical machine learning, which is a problem in their ability to effectively uncover hidden fraudulent trends and keep growing with the ever-evolving fraud strategies. Graph Convolutional Networks(GCNs) and Convolutional Neural Networks(CNNs) identify relational patterns in graph-structured data and extract spatial characteristics from transactional sequences, improving fraud detection. The proposed Graph-based Convolutional Fraud Detection (GCFD) methodology combines GCNs and CNNs to provide a sophisticated fraud detection system with improved accuracy. Company operations are graphed with accounts as nodes and financial transactions as edges; therefore, tracking transactions and regional fraud tendencies is essential. CNNs extract spatial relationships in transaction embeddings for fraud classification, whereas GCNs learn topological structures in transaction data. The results demonstrate that the GCFD model outperforms the current approach in identifying fraudulent transactions. It can detect intricate fraud rings that are missed by rule-based methods. According to the experimental data, GCFD outperforms both Random Forest and conventional anomaly detection models, with a 96.5% fraud detection accuracy and an F1-score of 94.2%. Furthermore, the model improves its reliability for real-world applications by reducing false positives by 23%. The suggested model is a flexible and resilient fraud detection system that dramatically enhances business transaction security.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Detecting Financial Fraud and Anomalies in Corporate Transactions with Convolutional Neural Networks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月25日

查看全文
  • Muqiao Cai

摘要

Financial fraud in business environments has evolved in response to the growth of digital transactions, posing a more significant threat to data security and resulting in substantial economic losses. The primary issue with current fraud detection systems is that they are either overly rule-based or too reliant on classical machine learning, which is a problem in their ability to effectively uncover hidden fraudulent trends and keep growing with the ever-evolving fraud strategies. Graph Convolutional Networks(GCNs) and Convolutional Neural Networks(CNNs) identify relational patterns in graph-structured data and extract spatial characteristics from transactional sequences, improving fraud detection. The proposed Graph-based Convolutional Fraud Detection (GCFD) methodology combines GCNs and CNNs to provide a sophisticated fraud detection system with improved accuracy. Company operations are graphed with accounts as nodes and financial transactions as edges; therefore, tracking transactions and regional fraud tendencies is essential. CNNs extract spatial relationships in transaction embeddings for fraud classification, whereas GCNs learn topological structures in transaction data. The results demonstrate that the GCFD model outperforms the current approach in identifying fraudulent transactions. It can detect intricate fraud rings that are missed by rule-based methods. According to the experimental data, GCFD outperforms both Random Forest and conventional anomaly detection models, with a 96.5% fraud detection accuracy and an F1-score of 94.2%. Furthermore, the model improves its reliability for real-world applications by reducing false positives by 23%. The suggested model is a flexible and resilient fraud detection system that dramatically enhances business transaction security.