КЛАСТЕРИЗАЦІЯ ОЗНАК МЕРЕЖЕВИХ АТАК В ЗАДАЧАХ АНАЛІЗУ ЗАХИЩЕНОСТІ ІНФОРМАЦІЇ

Abstract

The paper proposes an algorithm with self-learning elements for intrusion detection systems, as well as an improved clustering technique which is recorded by the data system concerning information security events. The proposed approaches differ from those known using an entropy approach allowing data to be presented as homogeneous groups, moreover, each such group (or cluster) may correspond to predetermined parameters. The proposed solutions relate to the possibilities of assessing dynamic dependencies between clusters characterizing the analysed classes of invasions. The studies have found that in case of manifestation of new signs of information security events, the corresponding scale changes and describes the distances between clusters. A computational experiment was conducted to verify the operability and adequacy of the proposed solutions. During the computational experiment, it has been found that step-by-step calculation of parameters of informative characteristics of network attacks allows to form sufficiently informative cluster structures of data having characteristic attributes. These attributes further become the basis for the knowledge base of intelligent network attack detection systems. Dynamic dependencies between clusters are calculated allowing for a sufficiently accurate definition of the many information security events that can become the source data for further automatic assessment of current threats extent detected by attack detection systems. The methodology and algorithm presented in the paper for clustering the signs of network attacks, in our opinion it is simpler for software implementation than existing analogues.