If I have a regression problem where I try to estimate the value of $y$ as function of $x_1 \dots x_d$: $$ y = f(x_1,\dots,x_d) $$

using a Boosted Regression Tree or a Random Forest Regression, is it possible to estimate the probability distribution of the output value (instead of only an estimate of $y$) :

$$ p(y\ |\ x_1,\dots,x_d) $$

from the distribution of the outputs of the individual nodes ? (i.e. for fixed $x_1,\dots,x_d$ I'm interested in having a probability distribution for $y$ from which one then could obtain the mode, mean and also the width which is an indication of 'how well' this value is predicted).


1 Answer 1


Yes, see, for example:

Meinshausen, Nicolai. "Quantile regression forests." The Journal of Machine Learning Research 7 (2006): 983-999.

  • 3
    $\begingroup$ Welcome to the site, @AJC. Would you mind expanding on your answer a little? EG, it would help readers if you could give readers a couple sentence overview of the paper & the answer contained therein. $\endgroup$ Mar 16, 2013 at 21:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.