Analysis of Network Intrusion Detection Based on Semi-Supervised and SS-DGM

Abstract

The rapid advancement of technology has made network security a hot topic of concern for researchers worldwide. Therefore, to improve the accuracy and real-time response capability of network intrusion detection systems, and to effectively detect and analyze network attack forms in complex network environments, starting from the drawbacks of existing network intrusion detection methods and the vulnerable attack modes of networks, this study first introduces variational auto-encoder to improve the semi-supervised intrusion detection algorithm. Subsequently, the labeled dataset is expanded using K-means clustering algorithm and collaborative training algorithm. Finally, a novel network intrusion detection model based on multi-level semi-supervised intrusion detection algorithm is proposed. The experimental outcomes denoted that the model achieved the highest classification accuracy of 93.08%, 92.12%, 91.22%, and 96.38% for four common types of network attacks: denial of service attacks, unauthorized access from remote machines, unauthorized access attacks by ordinary users to local superuser privileges, and cross-site scripting attacks, respectively. The results of the network attack confusion matrix indicated that the proposed model could detect and classify all types of network attacks, with excellent detection applicability and classification efficiency, and its scores were all above 60 points. In addition, the running time of the proposed model was only 25.64 seconds. From this, the proposed method effectively improves the dynamic adaptability and accuracy of network intrusion detection, providing an efficient and accurate solution for network security defense.

Keywords

• Semi-supervised
• SS-DGM
• variational auto-encoder
• clustering
• intrusion detection