GAN-Based Detection and Mitigation of Cyber-Attacks: A Comparative Study with Existing Techniques

Abstract

In present scenario network congestion problem occurs during data transferring and receiving on any networks. The various reason of this data flow is going slow due to various kinds of cyber-attacks respectively. The cyber-attacks work internally and damage the server system. it is very harmful to run system with occurring of cyber-attacks in network. this present paper presents the survey on cyber-attacks detection using machine learning techniques along with Jupiter and WEKA simulation tool. The various kind of cyber-attacks mention in this paper like DOS, TCP/IP attacks, flood attacks, UDP attacks, ICMP attacks, U2R, R2L, DDOS attacks, probe attacks along with detection using machine learning techniques and novel approach.  The machine learning techniques perform generalise data set on the other hand data mining techniques perform specific data set and detection of cyber-attacks the objectives of cyber-attacks detection and categories along with simulation process also elaborate.Building on identified research gaps, the paper proposes a hybrid Generative Adversarial Network plus Deep Learning (GAN+DL) framework in which the GAN generator synthesizes realistic minority-class and zero-day-like TCP/IP, UDP and ICMP attack traffic to balance training data, while a deep learning discriminator/classifierperforms high-accuracy detection, and a coupled mitigation module automates real-time response actions. Finally, an evaluation and comparison methodology is outlined to validate the proposed technique against existing detection and mitigation approaches using standard benchmark datasets. This PRISMA-based review of 40 IDS papers shows deep learning (92–98% accuracy) outperforms ML (78–85%) but needs more compute. Key gaps: high false alarms, poor U2R/R2L detection, no adaptability, and black-box issues. Offers a comparison framework for attack- and resource-aware IDS selection.