Optimization of classifier parameters when processing statistical characteristics of network packet metadata

Purpose of research. The article considers the possibility of increasing the probability of correct authentication of a remote message source based on the analysis of metadata of the network packets it generates. The purpose of this purpose is to develop a method for classifying authentic network packets based on the analysis of statistical characteristics of the packet arrival time and to optimize the classifier parameters to achieve maximum accuracy in determining authentic packet sequences. Methods. The study applies methods of analyzing high-order moments of interpacket intervals, as well as logistic regression for classifying packets. The parameters of excess and asymmetry calculated based on samples of time intervals formed by the arrival of packets are used. A classifier based on minimizing the distance from pairs of values (asymmetry and excess coefficients) to a parabola corresponding to the Poisson distribution is developed. Results. Samples with a power of 10 4 with calculated pairs of excess and asymmetry coefficients were formed. The obtained results show that for the maximum possible classification accuracy (82-84%), the optimal parabola parameters are: a ≈ 1.0, c = 8–9. ROC curves were analyzed for different sets of parameters, which confirmed the linearity of the dependence of the proportion of true positive results on the proportion of false positives. Conclusion. The results of the study confirmed the possibility of increasing the reliability of network packet authentication by using high-order moments of time interval data, which demonstrates the effectiveness of the proposed method. The main conclusions include the need for careful tuning of the classifier parameters to optimize the authentication process. Since the proposed method exhibits high sensitivity to changes in distributions, this opens up new directions for further research in the field of wireless network security.