A Novel Approach for Real-Time Server-Based Attack Detection Using Meta-Learning

Modern networks are crucial for seamless connectivity but face various threats, including disruptive network attacks, which can result in significant financial and reputational risks. To counter these challenges, AI-based techniques are being explored for network protection, requiring high-quality datasets for training. In this study, we present a novel methodology utilizing a Ubuntu Base Server to simulate a virtual network environment for real-time collection of network attack datasets. By employing Kali Linux as the attacker machine and Wireshark for data capture, we compile the Server-based Network Attack (SNA) dataset, showcasing UDP, SYN, and HTTP flood network attacks. Our primary goal is to provide a publicly accessible, server-focused dataset tailored for network attack research. Additionally, we leverage advanced AI methods for real-time detection of network attacks. Our proposed meta-RF-GNB (MRG) model combines Gaussian Naive Bayes and Random Forest techniques for predictions, achieving an impressive accuracy score of 99.99%. We validate the efficiency of MRG using cross-validation, obtaining a notable mean accuracy of 99.94% with a minimal standard deviation of 0.00002. Furthermore, we conducted a statistical t-test to evaluate the significance of MRG compared to other top-performing models.