Forecasting Information Security in Information and Communication Systems Based on Recurrent Neural Networks

Nowadays, one of the ways to solve the problem of proactive information security is to implement the function of forecasting information security. The paper explores an approach to forecast information security in information and communication systems (ICS) under the influence of computer attacks, based on the use of recurrent networks. As forecast values, it is proposed to use individual characteristics of the processes of security violations and restoration of information security, single and complex indicators of its security. Expressions for calculating information security indicators are proposed. A forecast model of information security in ICS based on a recurrent LSTM network is presented. An example of forecasting a complex indicator of information security in ICS based on an LSTM network is presented. A comparative assessment of the accuracy of forecasting information security was carried out, which showed that the proposed approach makes it possible to increase the accuracy of forecasting information security compared to traditional statistical methods and to provide a time reserve for taking proactive measures in order to prevent a decrease in the information security level. In addition, the advantage of the proposed approach to forecasting information security based on an LSTM network is that it more adequately takes into account the complex dependencies of information security violation and restoration processes.