"Importance of Z-transformation in clustering"

imfaith

Hello every body;

I haven't understand the importance of use Z-transformation in clustering process (for example k-means). It is necessary to use this transformation and what is its impact ? this transformation don't change the result of clustering ?  There are several methods of transformation, what is the most suitable for the clustering process ? ???
Thank you very much.

wessel

Assuming you are using k-means uses squared distance.
Assuming you create a a new attribute: att_1b = att_1 * 100.
If you were to use att_1b instead of att_1 during clustering, distances in att1_b are far more importance, because of the scaling.
Typically you want all attributes have equal weight in the distance calculation, Z-transforming all your attributes allows you to do just this.

Best regards,

Wessel