1
$\begingroup$

In the randomForestSRC in R, we have the example

data(veteran, package = "randomSurvivalForest")
pt.train <- sample(1:nrow(veteran), round(nrow(veteran)*0.80))
veteran.out <- rsf(Surv(time, status) ~ ., data = veteran[pt.train , ])
veteran.pred <- predict(veteran.out, veteran[-pt.train , ])

How can I tell if my model predictions are are good using the veteran.pred results?

Improve this question
$\endgroup$

1 Answer 1

Reset to default
1
$\begingroup$

The library "rms" has rcorr.cens which computes Harrell's C-index and similar measures.

Edit:

C-index is generalization of AUC (area under ROC curve), basically it's estimating the probability that an individual (person or whatever you're observing) with longer survival time is more highly ranked by your model than an individual with shorter survival time, over all pairs of samples. Like AUC, 0.5 is random, closer to 1 is better. This is explained in Frank Harrell's book Regression Modeling Strategies, plus in the documentation for "rms". You can also compare two models using rcorrp.cens, so say which model has better discrimination over the same samples.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Can you say a little bit more about this index & the "similar measures" to explain how they will solve the OP's question? (This post is being automatically flagged as low quality because it is so short; it is more of a comment than an answer in its current form.) $\endgroup$
    – gung - Reinstate Monica
    Commented Jun 19, 2014 at 5:08

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.