2
$\begingroup$

I am a Machine learning newbie and i am trying my hands with a dataset which has 9 features and my aim is to figure out the optimal multi class classification model which fits my dataset.

I applied PCA on my dataset and reduced the dimensionality to 2 and now i can visualize my dataset and this is how it looks like on my training dataset.

enter image description here

I am not sure how to choose a best model when data is distributed in such a way and i am looking for suggestions on what techniques are out there.

Is Getting more data a viable next step?

$\endgroup$
2
  • 1
    $\begingroup$ Add a sample dataset (df.head()) maybe.., Engineer new features, And also check Nn's if you have sufficient data. $\endgroup$ – Aditya Nov 30 '18 at 6:53
  • 1
    $\begingroup$ Having a bit of your data is crucial, otherwise impossible to suggest anything. And on top of that, why you think from PCA you can judge your data spread and choose the model. It might that PCA is not able to project your data to lower dimension properly. At least try to look at 3D projection, the data might be separable in 3D space. $\endgroup$ – TwinPenguins Nov 30 '18 at 8:45
1
$\begingroup$

It is questionable whether doing the PCA and reducing the dimension to 2D was a good idea. You can try experimenting with the number of components in PCA or even trying some non-linear dimensionality reduction like LLE:

Locally linear embedding (LLE) seeks a lower-dimensional projection of the data which preserves distances within local neighborhoods. It can be thought of as a series of local Principal Component Analyses which are globally compared to find the best non-linear embedding.

$\endgroup$
0
$\begingroup$

my aim is to figure out the optimal multi class classification model which fits my dataset

That is impossible for two reasons: (1) there are many reasonable ways to define "optimal", (2) there are infinitely many models and currently there is no analytical way to get a better one (not even speaking about the best one)

Have a look at my master thesis, chapter 2.5 for ideas how to continue.

$\endgroup$

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.