1

I came across the post here which mentioned the following:

k-means assumes the variance of the distribution of each attribute (variable) is spherical

I wanted to understand what does it mean by "variance of the distribution is spherical". Could anyone please help me understand this?

Curious
  • 421
  • 1
    https://stats.stackexchange.com/search?tab=votes&q=spherical%20variance turns up explanations. The first I found is at https://stats.stackexchange.com/a/144896/919. – whuber Jun 08 '22 at 16:30
  • Thanks for the reference @whuber. Is there a less mathy and more intuitive approach to understand the meaning of variance of distribution being spherical? – Curious Jun 10 '22 at 10:48
  • Yes: this is techno-speak for assuming the attributes (modeled as random variables) are uncorrelated and share a common variance. That's all. When the joint distribution is Normal, isocontours of the density are indeed (hyper)spheres. That's relevant to k-means because it implies any cluster created through joint random variation around a single point would have a ball shape rather than a strongly ellipsoidal shape. – whuber Jun 10 '22 at 12:52

0 Answers0