HPForest: Continuous Target Under-Prediction

[bluesky]

I was wondering if there are any quick ways to resolve an under-prediction problem (continuous target variable with a large number of 0s) if we're using Proc HPForest? 

[alexchien · Accepted solution]

maybe you can try 1st predicting if the target is 0 or not (a classification model), and then 2nd predicting the target if the they are not 0. You need to exclude the 0 observations from the 2nd model. alex

[bluesky]

Thanks so much -- yes, I'd been thinking of a two-part / hurdle model sort of setup but had delayed going down that road in case there was a more straight-forward solution.