Identification of the message flow between two subscribers in multi-agent systems based on the analysis of its contextual characteristics

Objectives . The paper examines the problem of improving the accuracy of identifying the message flow between two subscribers in multi-agent systems. This is done by analyzing the contextual characteristics of the overall message flow in the communication channel. Situations may arise during the process of source identification and verification of authenticity in which authentication codes for two or more messages collide. One way to resolve such conflicts is to isolate the message flow between two subscribers by leveraging its unique statistical characteristics which differ from the characteristics of the general message flow within the system. The aim of the paper is to develop a method which reliably identifies the target message flow even in cases of authentication code collisions. Methods . The contextual characteristics of messages are used, in order to analyze and highlight patterns of agent behavior in the message flow. These characteristics include frequency of sending, message size, timestamps, and historical interaction data. The method involves the formation of statistical characteristics of the message flow between two agents in a multi-agent system, such as skewness and kurtosis, as well as distribution parameters for the number of messages sent between events from the target source along with their classification by means of logistic regression. Results . During experiments, the method developed has been found to demonstrate a Precision metric value in the range of 0.81–0.85. This is 40–50% higher than existing methods based on the analysis of inter-packet time intervals, indicating that 81–85% of the messages classified as belonging to the target source are actually such. ROC 2 analysis confirmed the high efficiency of the model and acceptable classification quality. Conclusions . The results of the study show that the use of contextual characteristics and statistical analysis enables the accurate identification of target flows in multi-agent systems with a total number of agents ranging from 70 to 110. This method can be used in low-bandwidth communication channels where it is essential to minimize the size of the transmitted batch header and computational costs associated with authentication procedures.