Skip to main content

Table 1 Parameters of a Random Survival Forest with suggested values

From: Random survival forests for dynamic predictions of a time-to-event outcome using a longitudinal biomarker

Parameter Description Suggested value
ntree how many subsamples should be performed 500-1000
sampsize proportion of the data should be selected for each subsample 63% of available data
samptype Sampling Done With Replacement (swr) or without replacement (swor) swr
nodesize number of unique cases in each terminal node 15
mtry how many variables should be included as candidates for splitting \(\sqrt {\text {\# variables}}\)
nsplit how many splits should be considered for each candidate variable 10
splitrule impurity measure which aims to create the two most dissimilar daughter nodes at each split “logrank”