Density K-means: A new algorithm for centers initialization for K-means

Abstract

K-means is one of the most significant clustering algorithms in data mining. It performs well in many cases, especially in the massive data sets. However, the result of clustering by K-means largely depends upon the initial centers, which makes K-means difficult to reach global optimum. In this paper, we developed a novel algorithm based on finding density peaks to optimize the initial centers for K-means. In the experiment, together with our algorithm, nine different clustering algorithms were extensively compared on four well-known test data sets. According to our experimental results, the performance of our algorithm is significantly better than other eight algorithms, which indicates that it is a valuable method to select initial center for K-means.