Semi-Supervised Intrusion Detection Method based on Adversarial Autocoder

In terms of the problems of low-grade accuracy and high-grade inaccurate alarm rate for intrusion detection algorithm on the basis of unsupervised learning, and high cost of training samples required by supervised algorithm, an intrusion detection algorithm ground on adversarial autocoder is proposed. This is a semi-supervised learning algorithm, which only needs a handful of labeled data in the training data set for training, and supports unlabeled data in the training data set, so as to improve the performance. Firstly, the autocoder reduces the dimensionality of the import data by withdrawing essential features as reference of latent variables; secondly, it uses the generative adversarial network to make the latent variables of the autocoder follow an arbitrary distribution for regularization; and finally, it uses the cross entropy loss of labeled data to achieve the classification of semi-supervised learning. Experimental results show that, compared with other algorithms, the proposed algorithm has certain advantages in detecting a limited number of labeled samples, which can achieve high accuracy and low false alarm rate, while reducing the demand for labeled data.