0
$\begingroup$

I am working on a project at a company where I have to make clustering/unsupervised model. The data I am working on is very sparse with high dimensions and after some research, I found out TruncatedSVD is good for sparse data. So I applied TruncatedSVD on my data and ran a few clustering(k-means and GMM) algorithms on it and then I used top2 and top3 featured from truncatedSVD transformation to visualize the data in 2d and 3d space but the clusters are not distinctively separated.

So I did some more research I found out that kernel PCA is good at finding the separation between the data but it might not be good for sparse data. So my question is Can I use kernel PCA after transforming the data with truncatedSVD?

Will this be a correct approach that I can try? or are there are any other suggestion on how to tackle this problem?

To give you an idea about the dataset My data includes user_ids(unique) and their corresponding click frequency on multiple webpages and the click frequency on topics also there are some more variables related to time like time different between session and number of distinct days the user has visited the website etc.

$\endgroup$
0
$\begingroup$

It's one of her many combinations that does not make a lot of sense when you bother to try to understand the methods.

SVD tried to generate a projection where there is some kind of linearity in the factors. KernelPCA is used when you know non-linear relationships exist in your data.

It's generally a bad idea to just stack method on method that you don't know well. It is, unfortunately, easier to get a good looking but wrong result than an okay looking sound result. You should rather spend that time to better understand your data, your problem, and the theory of the methods, to identify how to align these three. In the end, don't aim for a result, but for an explanation.

$\endgroup$

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.