Detection of network attacks by deep learning method

Abstract

In this paper, we discuss the possibility of using a neural network approach in solving the problem of detecting network attacks. The neural network is designed to solve the problem of classifying the transmitted traffic into not containing an attack and containing an attack. The difficulties of using a neural network in this problem are discussed. Difficulties are associated with the choice of significant information features for the formation of the date set and having as many training examples as possible for each type of attack. Solutions for the selection of significant information features are proposed. It consists in ranking the features in order of importance. A method and rules for ranking features are proposed. In the future, it is proposed to use only important features to train the neural network. The problem of uneven number of training examples for each type of attack is considered. It is proposed to preserve significant examples represented by small sample sizes by assigning weights to them. Experiments show the effectiveness of the proposals discussed in the paper.