PCA algorithm: from data projection into a new feature space to outlier detection

Asked Sep 15 '16 at 22:35

Active Sep 15 '16 at 22:55

Viewed 195 times

I arrived at a point where I have eigenvectors and eigenvalues, therefore I can compute the projection matrix. What would be the next computational steps for determining a new data is an outlier or not? Supposing I plotted my current data points into the new projection space, how would I determine a new data point belongs to the group or not? Thank you.

P.S. as a concrete example, let's take this tutorial in Youtube https://www.youtube.com/watch?v=_UVHneBUBW0 How to determine a new Cell is an outlier or not?

My guess, intuitively: project the new data point into the PC space. Then, either calculate the distance to the cluster center OR find the closest point and see if it's not too far. What else? And how (both intuitively and formally)

edited Sep 15 '16 at 22:55

asked Sep 15 '16 at 22:35

Madrugada

The video does not seem to be talking about outlier detection at all. – amoeba Sep 16 '16 at 00:18
Right, I mean taking the cell example from the video, how would it be continued with outlier detection? – Madrugada Sep 16 '16 at 02:35
See related question http://stats.stackexchange.com/q/71899/3277 – ttnphns Sep 16 '16 at 03:55

PCA algorithm: from data projection into a new feature space to outlier detection

0 Answers0