0
$\begingroup$

I am using randomForestSRC to fit a random forest of 1000 trees for a 3-class classification problem with 160 cases. I am observing insanely high OOB error rates, sometimes as high as 75%, which is worse than guessing. What could be causing this? My calls are of the form

rfsrc(formula, data=df, block.size=1, ntree=1000, importance=TRUE,  verbose=FALSE, splitrule="auc", nsplit=NULL, nodesize=1)
$\endgroup$
  • $\begingroup$ Your categories may be poorly described by your data, try a different algorithm. $\endgroup$ – user2974951 Aug 12 at 11:19
0
$\begingroup$

How many features does your input data have? Given that you only have a 160 cases this looks like it could be over-fitting. PS: This should be more useful as a comment but I cannot post any yet

$\endgroup$
  • $\begingroup$ Anywhere from 5 to a couple of dozen, but even the 5 case has an OOB error of 0.7. $\endgroup$ – user11130854 Aug 12 at 9:23
  • $\begingroup$ Are you using sensitive data? If you're not, and you're happy to share I'll try it on my end, see what's going on. $\endgroup$ – a gonzalez Aug 12 at 9:27
  • $\begingroup$ Thanks for the offer! It is highly sensitive patient data, unfortunately. $\endgroup$ – user11130854 Aug 12 at 9:33

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.