Annotation

Annotation: The article considers an original approach to clustering of single-dimensional streaming data, based on the principles of density based clustering. This allows to work in conditions of information noise in order to cut off outliers and uninformative anomalous data. To implement this approach, an algorithm was developed consisting of several functional blocks and involving the search for single-dimensional cluster boundaries using machine learning technologies, which effectively uses information about the appearance of new clusters, preserving only significant data elements, which has a positive effect on the requirements for computing resources. To further improve the efficiency of the proposed algorithm, an approach of adaptive splitting of data from the input stream into frames of various sizes with subsequent processing based on a heuristic approach that takes into account the features of single-dimensional feature space and the cumulative nature of information about the presence of clusters is applied. The resulting algorithm demonstrates high efficiency in terms of data processing speed and memory usage. Its computational complexity tends to be linear over time. The authors also managed to achieve high clustering quality indicators, evaluated according to the criteria of compactness and separability of clusters, which are universal for any clustering algorithms based on the density of data distribution in the feature space. These advantages were confirmed by an experiment on 20 sets of test data, the results of which are also presented in the framework of this work. The presented algorithm occupies a rare niche of algorithms for clustering streaming data in conditions of information noise, optimized for working with one-dimensional data. Individually, each of the tasks of clustering streaming data and clustering one-dimensional data has been considered by the scientific community for quite a long time, however, their totality remains without due attention, despite the obvious benefits, for example, for solving problems of searching for stable states or clearing anomalous and noise values when analyzing one-dimensional signals, sensor readings, etc.