Neuroimmune method of data compression in the analysis system of information security incidents

Abstract

The article presents the development results of the neuroimmune method of data compression intended for further testing in an information security incident analysis system. The analysis of classical methods and machine learning methods used for data compression with and without losses was carried out. Options for neural network architectures to provide vector compression are considered. A method is proposed for combining a hybrid artificial immune system with a convolutional neural network of the Bottleneck-type architecture by using intelligent optimization and classification methods developed in previous research, including those that include a modified genetic dueling algorithm. The efficiency of the obtained hybrid compression approach with a classical convolutional neural network and a Bottleneck-type multilayer feedforward neural network has been evaluated. The degree of compression and accuracy of data decompression have been chosen as performance criteria. To conduct a comparative analysis of the considered methods, there has been developed a software package that implements a module for collecting and storing data in the information security incident analysis system, intended for further use in preparing data in the task of analyzing and correlating information security events. As a result of the given research, the developed neuroimmune compression method was proposed as the most effective approach to data compression, which showed, in a comparative analysis with its constituent algorithms, the best decompression accuracy at similar compression levels. The proposed method can be adapted for use in systems for compressing and storing not only text, but also video information, and can improve the quality of recognition by an incident analysis system when classifying images of security events.