A GAN and Feature Selection-Based Oversampling Technique for Intrusion Detection

Table 1

Design rationale of different oversampling techniques.

Technique	Design rationale
Random oversampling	Minority samples are randomly selected and replicated to increase the number of samples
SMOTE	Samples are generated by interpolation between each minority sample and its surrounding minority samples
ADASYN	As with SMOTE, ADASYN generates new samples by interpolation; the difference is that the number of new samples that need to be synthesized for each minority sample is determined by the density of majority class instances around it
K-means SMOTE	K-means SMOTE will first cluster the data into multiple clusters; different samples are then generated for the clusters’ density, with smaller densities generating a more significant number of samples
G-SMOTE	G-SMOTE generates synthetic samples in a geometric region of the input space, around each selected minority sample