Bisecting K-means algorithm based on density dividing principle

Abstract

The invention relates to a bisecting K-means algorithm based on a density dividing principle. Through the bisecting K-means algorithm, a data set is divided into two clusters. Furthermore by means of a dbscan algorithm, the number of sub-clusters which can be obtained through dividing all the clusters is obtained. The cluster with largest number of sub-clusters is further divided through the bisecting K-means algorithm. Clustering is finally finished through continuous iteration. The bisecting K-means algorithm has advantages of overcoming a defect of low sensitivity to cluster shape in an sse minimum dividing principle, more scientifically selecting the cluster to be divided, effectively improving clustering accuracy, furthermore realizing requirement for comparison of the number of sub-clusters in each cluster on the condition of same neighbor radius and number-of-points threshold, preventing clustering accuracy reduction caused by improper neighbor radius and number-of-points threshold setting, overcoming a defect of incapability of eliminating noise points in partitioning clustering, and realizing high realistic meaning.

Description

technical field [0001] The invention relates to the technical field of data clustering, in particular to a binary K-means clustering method based on a density division criterion. Background technique [0002] Clustering, also known as cluster analysis, refers to assigning data to different clusters, so that the data difference in the same cluster is as small as possible, and the data difference in different clusters is as large as possible. Clustering algorithm is an unsupervised learning method, which is widely used in data mining, recommendation system and other fields. Clustering methods can be roughly divided into partition clustering, hierarchical clustering, density clustering, grid clustering and model clustering. [0003] Among them, partition clustering is an easy-to-understand clustering method and the most common clustering algorithm. The famous k-means algorithm is a typical one. The Kmeans algorithm is widely used because it is easy to understand and has low t...

AI Technical Summary

Problems solved by technology

DBSCAN is a representative algorithm based on density clustering. It can find clusters of any shape in the spatial database, and can find and remove noise points. However, it is very sensitive. If it is not set properly, it will cause a decrease in the quality of clustering.

Images