High-order fuzzy clustering algorithm based on multikernel mean shift

This study proposes a method of constructing multikernel space to ensure the integrity of the original data in which the multikernel space aims to reduce the computational complexity of multidimensional data and is suitable for the processing of relational data. The high-dimensional samples of the original space are therefore mapped into a high-dimensional kernel feature space to obtain the inner product. However, when the dimensions of the feature space for multikernel is extremely high or even infinite, the inner product is difficult to calculate directly. To overcome these limitations, this study further proposes a high-order fuzzy clustering (HoFC) algorithm called multikernel mean shift (MKMS-HoFC), which incorporates mean shift based on multikernel space to divide the data and expand the original dimension into multiple new dimensions in the high-dimensional kernel feature space. The MKMS-HoFC initially maps the input points into a high-dimensional feature space of the multikernel and constructs a separating hyper-plane that maximizes the margin among multiple clusters in this space. The multikernel then finds the optimal hyper-plane by HoFC. This method iteratively searches for the densest regions of the sample points in the feature space and improves the clustering performance by using the multidimensional commensurability of HoFC. Real datasets are used to analyze the quality of clustering. Experimental results and comparisons demonstrate the excellent performances of MKMS-HoFC with its effectiveness in practice. (C) 2019 Elsevier B.V. All rights reserved.