Since the expected value that minimises the error at each leaf is used for prediction does this mean that the forecasted values can never be smaller than the smallest value in the dataset or bigger than the largest?
-
$\begingroup$ In short, yes, your assertion is correct. See Random forest regression not predicting higher than training data $\endgroup$ – Firebug Sep 30 '16 at 18:04
-
$\begingroup$ Thanks @Firebug, good to know I was on the right track but unfortunately it makes RF unsuitable for my problem. $\endgroup$ – Warren O'Neill Oct 3 '16 at 8:24