1
$\begingroup$

I'm looking for a way to perform multiple simultaneous quantile regression.

I read that the quantile crossing phenomenon is a common issue for that type of tasks, and I must avoid it.

So my question is : does Quantile Forest Regression prevent quantile crossing, or not? If not, could you a better suited method?

Thanks in advance!

$\endgroup$
1
$\begingroup$

Yes, quantiles of QFR never cross.

Accordingly to the paper of Meinshausen, quantile regresion forest works as follows:

fit: just fit an ordinary ransom forest

predict: for an observation $x$, take all the trainging samples that shared leaves with $x$, and return their quantiles.

Because the algoritm always returns "real" quantiles (although maybe of somehow "imaginary" distribution), they are ordered properly for each possible test sample.

$\endgroup$

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.