Markov?GAN: Markov image enhancement method for malicious encrypted traffic classification

The rapidly growing encrypted traffic hides a large number of malicious behaviours. difficulty collecting and labelling makes the class distribution dataset seriously imbalanced, which leads to poor generalisation ability classification model. To solve this problem, new representation learning method in its diversity enhancement model are proposed, uses images represent samples. First, is transformed into Markov images. Then, maximisation Markov-GAN based on Simpson index designed generate Finally, balanced image set sent CNN for classification. Experimental results show that proposed can predict whole space with only few original And accuracies under different imbalance degrees significantly improved, all over 90%. enhanced effectively alleviate performance deviation caused by network depths. Even an ordinary has almost same effect as VGG13 VGG16. Compared other data methods, needs balance transform domain dataset, lightweight, easy train stronger amplification ability.