5
$\begingroup$

I have a data set that looks like this:

data set

I would like to estimate a relationship between x-values and the corresponding 5% extreme y-values, something that might look like that :

regression on extreme values

Do you have an idea of an algorithm that might help me for this ? I thought about labelling the extreme values for later finding a separating hyperplane, but I have no clue on how to label these "extreme values" (I cannot just take the 5% lowest and highest values as all these would end up in the same region).

Thanks for your ideas !

Improve this question
$\endgroup$
1
  • $\begingroup$ Welcome to AI! You many want to consider adding an addendum to explain that the overall problem is not restricted to two dimensions. :) $\endgroup$
    – DukeZhou
    Commented Apr 27, 2018 at 16:16

1 Answer 1

Reset to default
2
$\begingroup$

I don't know of a pre-canned algorithm but I would just sweep on the angle from zero to ninety degrees with a triangular region and count the points. For each step in the sweep, record the angle and the count. When the sweep is done you will have an array of angles with bin counts and then you can convert to percentage of total count. You will have to figure out a resolution of the triangle's angle so that it is fine enough to capture less than 5% of the points.

enter image description here

Improve this answer
$\endgroup$
2
  • 1
    $\begingroup$ Thanks for your idea ! I forgot to mention it but I used a 2-dimensional example only because it was easier to explain. I would like to do this in higher dimensions (more predictors), which makes it harder... $\endgroup$
    – olafgrossebaf
    Commented Apr 27, 2018 at 12:49
  • $\begingroup$ @user15281 I'd upvote this answer since you didn't specify n dimensions in the question, and it's useful to see 2D approaches in building toward n dimensional approaches. $\endgroup$
    – DukeZhou
    Commented Apr 27, 2018 at 16:15

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .