A Clustering algorithm of high-dimensional data based on sequential psim matrix and differential truncation

Abstract

For high-dimensional data, the failure in distance calculation and the inefficient index tree that are respectively derived from equidistance and redundant attribute, have affected the performance of clustering algorithm seriously. To solve these problems, this paper introduces a clustering algorithm of high-dimensional data based on sequential Psim matrix and differential truncation. Firstly, the similarity of high-dimensional data is calculated with Psim function, which avoids the equidistance. Secondly, the data is organized with sequential Psim matrix, which improves the indexing performance. Thirdly, the initial clusters are produced with differential truncation. Finally, the K-Medoids algorithm is used to refine cluster. This algorithm was compared with K-Medoids and spectral clustering algorithms in two types of datasets. The experiment result indicates that our proposed algorithm reaches high value of Macro-F1 and Micro-F1 at the small number of iterations.