Аннотация
In the context of the rapid growth of network traffic volumes and the complexity of corporate information systems (IS), the development of effective methods for data analysis and clustering is of particular relevance. Modern approaches to network traffic processing require a comprehensive consideration of numerous parameters and characteristics, which necessitates the improvement of existing cluster analysis methods. The paper proposes an improved approach to ensemble clustering of network traffic and a method for constructing a consistent similarity matrix for integrating the results of different clustering algorithms based on exponential dependence in order to enhance the differences in the weights of the algorithms, which significantly increases the accuracy of the final clustering. The software system implemented in the Python language as part of the study combines several clustering methods, which allows achieving significantly greater stability of the results through the use of a consensus approach, the effectiveness of which has been confirmed by the results of large-scale computational experiments. During the research and computational experiments, it was demonstrated that the key advantage of the developed approach is increased resistance to outliers compared to traditional methods (KMeans, DBSCAN) used for cluster analysis of the company's network traffic, as well as more balanced clustering for complex multidimensional data. A dependence for calculating algorithm weights using the exponential function is also proposed, which allows for a comprehensive approach to integrating the results of different clustering methods. The developed software solution significantly expands the methods of network traffic analysis and provides an effective practical toolkit for increasing the productivity of corporate information systems. The proposed approach can be successfully adapted to solve a wide range of data analysis problems that require processing large volumes of multidimensional information.