Combating Imbalance in Network Traffic Classification Using GAN Based Oversampling

Generative Adversarial Networks (GANs): A Survey on Network Traffic Generation

“…We now summarize these, concentrating mainly on their structure. [48] This recent development in GANs addresses the problem that typical Internet traffic has very different proportions of traffic from different applications, leading machine learning training to be dominated by the most commonly seen type.…”

“…ITCGAN, inspired by the triple-GAN [48] framework, is structured to include three modules as shown in Fig. 2.…”

“…The last of these comprises the GAN framework that includes 𝐺, 𝐷 and a Classifier (𝐶𝑙). 𝐺 is designed with Weight Generation Units (𝑤𝐺𝑈) , which each correspond to a minority class and learn a latent space's conditional mapping 𝑔 𝑖 to vector 𝑤 𝑖 = 𝑔 𝑖 (𝑧/𝑖) of weights 𝑁 𝑖 [48]. Unlike [26] that trains so as to map a uniform random distribution that is similar to 𝑃 𝑑𝑎𝑡𝑎 (𝑥) to target data, 𝐺 is trained to learn and synthesize minority samples that fit the original distribution even though this differs from 𝑃 𝑑𝑎𝑡𝑎 (𝑥).…”

“…𝑝 𝑖 𝑑 and 𝑝 𝑖 𝑔 respectively indicate the real and synthetic conditional probability distributions of class 𝑖, and 𝑁 𝑛 − 𝑁 𝑖 is the class size. ITCGAN attempts to minimize (7) to fool 𝐷, and maximize ( 8) and ( 9) to enable 𝐶𝑙 predict the synthetic samples as real labels [48].…”

“…The GAN architecture incorporates facilitation of correct classification of the imbalanced set while serving as a constraint to guide 𝐺 during training, and also providing an indication of successful generation thereby eliminating the need to focus on training convergence [48].…”