K-means is another clustering method. It is not hierarchical which gives it less rigidity and more flexibility. It uses the same measure as the Ward method but its process is different. The users have to nominate how many clusters they want to be formed. The computer then puts that number of centroids, in our case in the 28 dimensional space, randomly (though in SPSS 17 it is always the same places as the results are always the same if the order of the forces is the same). The computer, through a number of tries or iterations, attempts to achieve the Sum of Squares minimisation within clusters and maximisation between clusters discussed above until it can do no better. You can see the non-hierarchical flexibility with Cleveland (in the North East) starting purple, going green, then going back to purple; something that is not possible using a hierarchical method.
Search This Blog
Wednesday, 11 August 2010
Explaining clustering to non- mathematicians and non-geographers 6
K-means is another clustering method. It is not hierarchical which gives it less rigidity and more flexibility. It uses the same measure as the Ward method but its process is different. The users have to nominate how many clusters they want to be formed. The computer then puts that number of centroids, in our case in the 28 dimensional space, randomly (though in SPSS 17 it is always the same places as the results are always the same if the order of the forces is the same). The computer, through a number of tries or iterations, attempts to achieve the Sum of Squares minimisation within clusters and maximisation between clusters discussed above until it can do no better. You can see the non-hierarchical flexibility with Cleveland (in the North East) starting purple, going green, then going back to purple; something that is not possible using a hierarchical method.
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment