Improved initial clustering center selection algorithm for K-means

Abstract

To overcome the deficiencies of traditional K-means algorithm whose clustering effect and stability are easily affected by the initial clustering centers, this paper proposes an initial clustering center selection algorithm based on Max-Min criterion and FLANN. The algorithm firstly identifies K farthest objects, and then finds out the k nearest neighbors of K objects respectively. Finally, take the center of each k nearest neighbor object as the initial clustering center. The experiment shows that the initial clustering center selection algorithm based on Max-Min criterion and FLANN possesses higher accuracy and stability than selecting the initial clustering centers randomly, selecting the initial clustering centers density-based, selecting the initial clustering centers based on intelligent algorithm.