Scholarlink

<p>Financial statement fraud poses a significant threat to enterprise digital transformation as organizations increasingly rely on computerized financial systems, cloud accounting, and AI-driven financial decision-making. To address this challenge, this research introduces FinGuardNet, a Deep Learning (DL)–based fraud detection model designed to enhance fraud identification in financial statements. FinGuardNet integrates DenseNet for hierarchical feature extraction, Temporal Convolutional Networks (TCN) for sequential fraud pattern recognition and Transformer-based Self-Attention Mechanisms for feature weighting. Through collaborative fraud detection without exposing sensitive financial data, Federated Learning (FL) improves secure fraud detection among companies. The system is deployed via a Google Cloud Function, ensuring real-time fraud risk assessment and seamless enterprise integration. Experimental results demonstrate the model’s 99.60% accuracy, 99.80% precision, and a 0.99 ROC-AUC score, significantly improving fraud detection efficiency while minimizing false positives. With an inference time of 2.5 ms per transaction and 95% regulatory compliance, FinGuardNet ensures high-speed fraud identification while maintaining security and scalability. This research contributes to the advancement of AI-driven fraud detection by offering a robust, adaptive, and future-ready solution for financial fraud risk management in digital enterprises.</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

Research on fraud identification of financial statements in enterprise digital transformation based on deep learning algorithm

  • 期刊论文

  • 发表日期：

    2025年6月25日

查看全文
  • Jinxin Wang

摘要

Financial statement fraud poses a significant threat to enterprise digital transformation as organizations increasingly rely on computerized financial systems, cloud accounting, and AI-driven financial decision-making. To address this challenge, this research introduces FinGuardNet, a Deep Learning (DL)–based fraud detection model designed to enhance fraud identification in financial statements. FinGuardNet integrates DenseNet for hierarchical feature extraction, Temporal Convolutional Networks (TCN) for sequential fraud pattern recognition and Transformer-based Self-Attention Mechanisms for feature weighting. Through collaborative fraud detection without exposing sensitive financial data, Federated Learning (FL) improves secure fraud detection among companies. The system is deployed via a Google Cloud Function, ensuring real-time fraud risk assessment and seamless enterprise integration. Experimental results demonstrate the model’s 99.60% accuracy, 99.80% precision, and a 0.99 ROC-AUC score, significantly improving fraud detection efficiency while minimizing false positives. With an inference time of 2.5 ms per transaction and 95% regulatory compliance, FinGuardNet ensures high-speed fraud identification while maintaining security and scalability. This research contributes to the advancement of AI-driven fraud detection by offering a robust, adaptive, and future-ready solution for financial fraud risk management in digital enterprises.