A clustering method based on near neighbor density and manifold distance

Abstract

A clustering method based on near neighbor density and manifold distance comprises the following steps: 1) calculating the weight of each feature according to improved information entropy); 2) calculating the near neighbor density of each sample accord to the weighted Euclidean distance, and selecting a center point accord to the nearest neighbor density; 3) calculating the Euclidean distance of each sample in the data set obtained in the step 2, and constructing an adjacency graph; 4) calculating the manifold distance between every two vertices in the adjacency graph to form a manifold distance matrix; 5) selecting k initial clustering centers and classifying each point into a cluster represented by a clustering center with the smallest distance from the manifold; and 6) renewing the cluster center, then repeat step 5 until that cluster center is no longer changes or reaches the upper limit of iteration times. The invention provides a clustering method with high algorithm running efficiency and good clustering precision through the abovementioned method.

Description

technical field [0001] The invention relates to a clustering method, in particular to a clustering method based on neighbor density and manifold distance. Background technique [0002] With the development of information technology and Internet of Things technology, modern industry has accumulated a large amount of data over time. However, due to the mutual influence and interconnection of various variables in the production process, the distribution of these data in the sample space is complex. Industrial big data contains great value, and how to improve the availability of industrial big data and mine value from complexly distributed industrial big data has become a research hotspot. [0003] As an important data mining method, clustering can divide a data set into several clusters whose intra-class objects are as similar as possible and the inter-class data objects are different, so as to discover potential data patterns and internal connections in the data set. Most clu...

AI Technical Summary

Problems solved by technology

Euclidean distance cannot accurately describe it, and clusters of arbitrary shapes cannot be found

In addition, most scholars regard all attributes of the data as equally important when conducting cluster analysis research, that is, each attribute is set to the same weight during algorithm design, and does not consider the impact of data mining analysis due to unequal weights of data attributes. impact of results

Images