Clustering Algorithm Based on Feature Space Partitioning

In clustering problems, there are problems associated, in particular, with outliers and linear inseparability of data. A new approach to robust clustering is proposed based on recursive division of the feature space into hypercubes (cells). In the process of analyzing cell densities and their homogeneities, cells merge into clusters. The algorithm is quite resistant to outliers and is capable of separating linearly inseparable points. An algorithm for robust clustering of linearly inseparable points, its software implementation, as well as test results on classical datasets are presented. The advantages and disadvantages of the proposed method, as well as possible ways of its development are discussed.