Is it possible to use t-SNE, PCA or UMAP to find separating hyperplane?
Assume we have data points in high dimensional space and we want to phase separate it into two sets of points?
Is there a way to use t-SNE, PCA or UMAP for such a task?
All the methods are basically manifold methods which are used to squeeze hyper dimensions to two or three dimensions with a certain amount of information loss. So whatever you "see" with these methods are not real and deceiving. You may see a separation of data points in 3D but when they are mapped onto the actual dimensions it may be complete rubbish.
By definition, these methods do not provide a way to find separating hyperplane. They just tell you some condensed information about the whole data. It is generally advised to use methods like t-SNE with caution.