Adaptation of the neural network model to the identification of the cyberattacks type "denial of service"

The article is devoted to the improvement of neural network technologies for detection of cyber-attacks of the "denial of service" type. It is shown that to detect such cyber-attacks, neuronet models based on a multilayer perceptron are used, whose main task is to determine the admissibility of deviations of the current values of the security parameters of computer systems and networks from values characteristic of normal conditions. At the same time, there is a significant shortage of the multilayer perceptron associated with a low accuracy in predicting the deviations of the safety parameters in the region of their minimum values. It is determined that the reason for this drawback is not sufficient adaptation of the mathematical support for the learning process of the multilayer perceptron to the dynamics of the safety parameters. It is suggested to increase the adaptation due to the application of the target functional of optimizing the values of the weight coefficients in the form of a quadratic reduced learning error. This target functional is integrated into the mathematical apparatus used to train the multilayer perceptron. The expediency of the developed solutions is confirmed experimentally.