Abstract:

The initial clustering centers of traditional k-means  are randomly selected, which results in unstable clustering results. To solve this problem, we propose an improved algorithm based on discrete quantity. In the proposed algorithm, all the objects are firstly regarded as a class and the two objects that have the maximum and the minimum discrete quantity respectively are selected from the cluster with the largest number of objects as the initial clustering centers. And then the other objects in the largest cluster are partitioned to the nearest initial clusters. The partition process is repeated until the cluster number is equal to the specified value k. Finally, as the initial clusters, the partitioned k  clusters are applied to the k-means  algorithm. We conduct experiments on several datasets, and compare the proposed algorithm with the traditional k-means  algorithm and max-min distance clustering algorithm. Experimental results show that the improved k-means algorithm can select unique initial clustering centers, reduce the times of iteration, and has stable clustering results and higher accuracy.

Key words: discrete quantity, k-means;clustering, clustering center