Ensemble survival models

The random forest package randomForestSRC will continue to be useful for creating the random forests associated with the survival data. In fact, the S of SRC in the package name stands for survival! The usage of the rfsrc function remains the same as in previous chapters, and we will now give it a Surv object, as shown in the following code:

> pbc_rf <- rfsrc(Surv(time,status==2)~trt + age + sex + ascites + 
+                 hepato + spiders + edema + bili + chol + albumin+
+                 copper + alk.phos + ast +trig + platelet + protime+
+                 stage, ntree=500, tree.err = TRUE, pbc)

We will find some of the basic settings that have gone into setting up this random forest:

> pbc_rf$splitrule
[1] "logrankCR"
> pbc_rf$nodesize
[1] 6
> pbc_rf$mtry
[1] 5

Thus, the splitting criteria is based on the log-rank test, the minimum number of observations in a terminal node is six, and the number of variables considered at random for each...