Abstract
Generative Adversarial Network (GAN) has turned out to be one of the most devastating types of ransomware attack detection models, with enhanced normalization technique being the order of the day in most sectors. GAN methods of ransomware detection tend to be problematic in data Preprocessing, input dataset, and behavioural logs. In this paper, a better modification of GAN-based ransomware attack detection is proposed, which adds a framework such as a generator (G), discriminator (D), and auxiliary classifier (C). To achieve a better data representation, the GAN is used with Enhanced Normalization, with the enhancement of the normalization that stabilizes and reduces the occurrence of the mode collapse problem. Empirical tests reveal that the model under consideration is an improvement over traditional GAN-based and Map, instance normalization, Layer normalization, and hybrid normalization. Classifier Integration comprises the increased computational cost of adversarial training, the need for sigmoid activation, learning rate, total objective function, training, and Optimization when deploying in real-time scenarios on limited systems. Future directions will center on the optimization of lightweight flavors of the proposed model, the incorporation of federated learning as a privacy-preserving detection mechanism, and cross-domain generalization as a means of increasing protection to previously unseen ransomware families.