Research Article

Oversampling Imbalanced Data Based on Convergent WGAN for Network Threat Detection

Table 1

The symbols are used in the formulas from (1) to (15) in Section 3.3.

No.Math symbolsMeaning of the math symbols

1The training set
2The size of the training set
3The majority class of the training set
4The size of the majority class in the training set
5The minority class of the training set
6The size of the minority class in the training set
7The imbalance ratio of the imbalanced data sets
8The oversampling rate of the minority class
9The false data generated by the generator of GAN
10The output data of the generator
11The output data of the discriminator
12The set of all possible joint distributions of the combination of and
13One joint distribution of all possible joint distributions of the combination of and
14The function of neural network model
15The Lipschitz constant
16The Wasserstein distance
17, The iterations in the training process of CWGAN
18, The training epoch in each iteration
19The training epoch in iteration
20The loss function of the discriminator
21The training stride in the training process of CWGAN
22The optimal solution of
23The optimal value of
24The Wasserstein distance between generated data and real data when the discriminator is convergent at the -th iteration.