The k-means clustering method is used in non-hierarchical cluster analysis. The goal is to divide the whole set of objects into a predefined number (k) of clusters. The criteria for such subdivision is normally the minimal dispersion inside clusters – e.g. the minimal sum of squares of the distances from the mean vector ( centroid) of the cluster. A direct rigorous solution to this problem requires testing of an impractically large number of data subdivisions. The k-means clustering is a fast heuristic method that provides a reasonably good solution, although not optimal.
For more details see the chapter in the XLMiner help.