K-Means Clustering

K-means (Lloyd, 1957; MacQueen, 1967) is one of the most popular clustering methods. Algorithm ?? shows the procedure of K-means clustering. The basic idea is: Given an initial but not optimal clustering, relocate each point to its new nearest center, update the clustering centers by calculating the mean of the member points, and repeat the relocating-and-updating process until convergence criteria (such as predefined number of iterations, difference on the value of the distortion function) are satisfied.

The task of initialization is to form the initial K clusters. Many initializing techniques have been proposed, from simple methods, such as choosing the first K data points, Forgy initialization (randomly choosing K data points in the dataset) and Random partitions (dividing the data points randomly into Ksubsets), to more sophisticated methods, such as density-based initialization, Intelligent initialization, Furthest First initialization (FF for short, it works by picking the...