The optimized K-means algorithms for improving randomly-initialed midpoints

摘要

In view of the traditional k-means randomly generated initial cluster centers approach proposed three kinds of adaptive optimization algorithm that are the nearest neighbor K-mean, extreme neighbor K-means and adaptive K-means. The nearest neighbor K-means is to ascertain the K group by searching weighted Euclidean nearest point in multidimensional space; and the extreme neighbor K-means is farthest nearest decision method; adaptive K-means is setting data into the matrix, then do normalization and dualization processing with the matrix, and calculate each vector dissimilarity to determine and weight correction Euclidean distance of initial center points. These 3 kinds of optimization algorithm improve the original K-means, improve the stability of the algorithm and accuracy, and each of them is suitable for different application space.