I'm a little confused about the GMM. I need to make clusterization. Actually, I have four dimensions : speed in Km/H, mileAge in Km, acceleration in m/s^2 and braking in m/s(^2). Each of those dimensions appears to be a gaussian. But, for each of this distribution, there is only one gaussian (this is not a mixture of gaussians).

As I understand at the beginnig, because all of my dimensions seems like a gaussian, it was perfect to apply the EM algorythm on it (used by the GMM). But, looking deeper, I start to think that I need to have several gaussian for each dimension.

So my question is : When we talk about "mixture of gaussian", it means to have several gaussians in the same dimension or a mixture of dimensions that contains one gaussian ?

  • $\begingroup$ In general GMM is a multiple Gaussians in multiple dimensions. Even if your single dimension (I guess you mean feature/attribute) shows a unimodal distribution, it can happen that there is a multimodal distribution when you consider all the features. $\endgroup$
    – Krrr
    Aug 29, 2017 at 11:14

1 Answer 1


A mixture of Gaussians can be defined as $f(x) = \sum_{i}^{N} p_{i} \exp\big(-\frac{1}{2}(x-\mu_{i})^T\Sigma_{i}^{-1}(x-\mu_{i})\big)$. So it is $N$ Gaussians. The dimension of $x$ and $\mu_{i}$ can be one or bigger than one.

  • $\begingroup$ So I should choose the GMM instead of an algorithm like Kmean ? $\endgroup$ Aug 29, 2017 at 12:17
  • $\begingroup$ Both algorithms are possible, you can check which one gives you better results. $\endgroup$
    – citronrose
    Aug 29, 2017 at 12:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.