An efficient grid-based clustering method by finding density peaks

Abstract

Clustering or categorizing an unprocessed data set is essential and critical in many areas. Much success has been published, which first needs to calculate the mutual distances between data points. It suffers from considerable computational costs, preventing the state-of-the-art methods such as the clustering method by fast search and find of density peaks (FSFDP, published in Science, 2014) from applying into real life (e.g., with thousands of data points). In this paper, an efficient grid-based clustering (GBC) method by finding density peaks is described. It keeps the advantage of the friendly interactive interface in the FSFDP, at the mean time, decreases enormously the computation complexity. The time complexity of the FSFDP is o(np(np − l)/2) while our method decreases it to o(np * sizeof (grid)), where np is the number of data points and the size of grid is always much smaller than np so that the time complexity of our approach is almost linearly proportional to np. The presented GBC method by finding density peaks was able to calculate the densities and categorize datasets within much less time, which makes the density-peak-based algorithm practical. By using the presented algorithm, it was possible to cluster high-dimensional data sets as well. The GBC method by finding density peaks was successfully verified in clustering several datasets, which are commonly used to test clustering algorithms in published articles. It turned out that the presented method is much faster and efficient in clustering datasets into different categories than the conventional density-based ones, which makes the proposed method more preferable.