A Cybersecurity Threat Recognition Framework Combining GAN Networks and Semi-Supervised Learning

Abstract

Abstract This paper delineates the types of threat identification in network security, designs the threat identification model architecture, analyzes the malicious code, and proposes a counter-defense strategy. The talk compares the GAN network model and semi-supervised learning technology, combines them in network security, and proposes a semi-supervised detection model utilizing GAN. Analyze the characteristics of URL network activity and design URL character encoding. Set experimental parameters and selected datasets to analyze the similarity between synthetic URLs based on GAN generators and real URLs and test the effectiveness of a GAN-based semi-supervised detection model for malicious URL recognition using different classifiers. Calculate the detection model’s classification accuracy on a mixed dataset and test its training fit. There is only a 6% difference in how well different classifiers can spot malicious URLs. This shows that the adversarial samples made by the GAN-based generator are similar to real URLs. The GAN-based semi-supervised detection model is capable of recognizing more web threats.