Random Forest Optimization

[Jonison]

Hello, all, I got a database with 2 outputs, y1 and y2.

The Random forest analysis was performed, and then for y1 the average square error of baseline fit statistics is 33.4 which is too big, and for y2 the errors are less than 0.1.

I tired a couple of tricks, e.g. increase max tress, variables to try etc., but the average square error has not improved. Would you please give some suggestions on how to improve the RF predictivity?

Many thanks

[sbxkoenk]

Hello,

To me, it's all in feature engineering. That's where the gain lies.

If you search for SAS and "feature engineering" or "feature building", you find a lot of hits on Internet.

Also, Model Studio has some nodes for it.

Thanks,

Koen