Method for determining optimum cluster number

Abstract

Discloses is a method for determining an optimum cluster number. The cluster effect of a data set is evaluated through an effectiveness indicator Q (C), and the cluster number corresponding to the minimum value of the effectiveness indicator Q (C) is the optimum cluster number. According to the method for determining the optimum cluster number, a new similarity measuring method is provided, all possible cluster partitions are generated in a bottom-up mode by combining hierarchical clustering, the effectiveness indicator value at the moment is calculated, a cluster quality curve regarding different partitions is established according to the effectiveness indicator value, and the partition corresponding to the extreme point of the curve is the optimum cluster partition. Repeated clustering on a large data set can be avoided, and the method does not rely on specific clustering algorithms. Experimental results and theoretical analysis both show that the method has good performance and feasibility, and computational efficiency can be improved greatly.

Description

technical field [0001] The invention belongs to the technical field of data mining and relates to a method for determining the optimal number of clusters. Background technique [0002] Most of the determination of the optimal number of clusters is carried out using an iterative trial-and-error process. On a given data set, different parameters (usually the number of clusters k) are used to run a specific The clustering algorithm divides the data set differently, and then calculates the effectiveness index values ​​of various divisions. By comparing each index value, the number of clusters corresponding to the index value that meets the predetermined conditions is selected as the optimal number of clusters. . In fact, there are several deficiencies in the trial-and-error process. First, the determination of the k value of the cluster number is difficult for users who lack rich experience in cluster analysis. The method of reasonable clustering number k; the second is that m...

AI Technical Summary

Problems solved by technology

In fact, there are several deficiencies in the trial-and-error process. First, the determination of the k value of the cluster number is difficult for users who lack rich experience in cluster analysis. The method of reasonable clustering number k; the second is that many indicators for testing the effectiveness of clustering have been proposed, and the main representatives are V xie Index, V wsj Indicators, etc.

Since these indicators are proposed based on a specific clustering algorithm, the method is greatly limited in practical application.

In addition, this method has poor computational efficiency for large-scale and complex-dimensional data sets.

Images