Cross-site scripting attack detection based on a modified convolution neural network.

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• Cross-site scripting (XSS) attacks pose a significant threat to network security.
• Effective detection and interception of XSS attacks are critical research areas.

Purpose of the Study:

• To propose a novel deep learning model, the modified ResNet block and NiN model (MRBN-CNN), for improved XSS attack detection.
• To enhance the capabilities of existing convolutional neural network architectures for network security.

Main Methods:

• Developed a convolutional neural network (CNN) integrating a modified ResNet block and NiN model (MRBN-CNN).
• Preprocessed URLs based on XSS attack script encoding syntax and semantics.
• Improved ResNet residual modules and utilized 1*1 convolutions to replace fully connected layers.
• Extracted features from multiple perspectives for comprehensive analysis.

Main Results:

• The MRBN-CNN model demonstrated superior performance and faster convergence compared to traditional machine learning and deep learning models.
• Achieved high detection rates, including 99.23% accuracy, 99.94% precision, and 98.53% recall.
• Significantly outperformed a baseline by approximately 75% in detection rate.

Conclusions:

• The proposed MRBN-CNN model offers a highly effective solution for detecting and intercepting XSS attacks.
• The novel architectural modifications and preprocessing techniques contribute to superior network security performance.
• This research advances the field of automated threat detection in cybersecurity.