Scholarlink

The complexity of modern web application security has become a big challenge due to SQL injections and cross-site scripting attacks. Traditional detection techniques, such as payload-based or request-based, result in higher false positives and limited capability to address new threats. This research aims to develop a novel deep learning-inspired framework using a Dual-Stage Hamiltonian Deep Neural Network (DS-HNN) and advanced feature extraction and optimization techniques. Pretrained architectures like VGGNet, UNet, ResNet, and DenseNet are used for feature extraction in high-dimension space. The Dung Beetle Optimization (DBO) algorithm is introduced to improve detection performance with better computational complexity. The DS-HNN model distinguishes vulnerable samples and refines predictions to reduce false positives. Extensive evaluation on SQL-XSS Payload, Testbed, and CSIC 2010 datasets shows high detection accuracy up to 99.89%, outperforming existing methods. The proposed method is efficient, scalable, and can be adapted to secure up-to-date web applications from evolving threats.

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

Deep Learning-Based Vulnerability Detection for Web Application Security: A Dual-Stage Hamiltonian Neural Network Approach for SQLi and XSS Attacks

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Sonkarlay J. Y. Weamie,
  • Kahsay Birhanu Tsadik,
  • Vinothkumar Kolluru

摘要

The complexity of modern web application security has become a big challenge due to SQL injections and cross-site scripting attacks. Traditional detection techniques, such as payload-based or request-based, result in higher false positives and limited capability to address new threats. This research aims to develop a novel deep learning-inspired framework using a Dual-Stage Hamiltonian Deep Neural Network (DS-HNN) and advanced feature extraction and optimization techniques. Pretrained architectures like VGGNet, UNet, ResNet, and DenseNet are used for feature extraction in high-dimension space. The Dung Beetle Optimization (DBO) algorithm is introduced to improve detection performance with better computational complexity. The DS-HNN model distinguishes vulnerable samples and refines predictions to reduce false positives. Extensive evaluation on SQL-XSS Payload, Testbed, and CSIC 2010 datasets shows high detection accuracy up to 99.89%, outperforming existing methods. The proposed method is efficient, scalable, and can be adapted to secure up-to-date web applications from evolving threats.