A Fuzzy Clustering Algorithm Based on K-means

Traditional K-means algorithm cannot get high clustering precise rate, and easily be affected by clustering center random initialized and isolated points, but the algorithm is simple with low time complexity, and can process the big data set quickly. This paper proposes an improved K-means algorithm named PKM. PKM is based on similarity degree among data points made by cumulated K-means, and get the final clustering partition via fuzzy clustering analysis (Transitive Closure Method), to make the precise rate of clustering higher, and reduce the effects made by isolated points and random clustering center, at the same time, can recognize isolated points better. Experiments with analog data and real data demonstrate its advantage.