ENSEMBLE OF CLUSTERIZATION ALGORITHMS WITH VARIOUS METRICS FOR ANALYSIS OF BANK NETWORK TRAFFIC

Abstract

The relevance of the study is due to the need to increase the accuracy of the analysis of bank network traffic, represented by heterogeneous data, including server logs, network connections, user behavioral data and traffic telemetry. In the context of increasing data volumes and the complexity of the structure of corporate networks of Ukrainian banks, traditional analysis methods lose their effectiveness, making the application of machine learning methods, including ensemble clustering algorithms, relevant. Modern research in this area focuses on the development of adaptive segmentation methods that use probabilistic models and dynamic weighting of algorithms to increase the accuracy of data partitioning. The paper proposes an ensemble KA method that uses variations of the K-means algorithm with different distance metrics. The method is based on a probabilistic model that takes into account latent classes and dynamically adjustable algorithm weights, and for the consistency of the partitioning, a coassociative matrix is ​​used that reflects the frequency of pairs of objects falling into one cluster. The novelty of the research lies in the development of a mechanism for adaptive weighting of ensemble algorithms based on quality metrics (ARI, NMI), which allows to increase the accuracy of clustering of bank network traffic. The proposed method compensates for the shortcomings of individual algorithms and takes into account the features of different types of data (packets, connections, users, devices). The practical significance of the work is confirmed by the possibility of applying the method to analyze banking network traffic, detect anomalies, optimize routing and increase cybersecurity of the banking network, and the developed approach will allow to take into account the complex data structure and adapt to dynamic changes in the bank's network environment.