rangerror

Confidence intervals for the random forest generalization error.

Paulo C. Marques F.

Pattern Recognition Letters, Volume 158, June 2022, Pages 171-175

devtools::install_github("paulocmarquesf/rangerror")

library(ranger)
library(rangerror)
library(MASS)
library(mlbench)

set.seed(42)

Regression (Boston housing)

rf <- ranger(medv ~ ., data = Boston, num.trees = 10^3, keep.inbag = TRUE)

sqrt(oob_ci(rf))

    2.5%    97.5%
2.764147 3.663428

Classification (Forensic glass)

data(Glass)

rf <- ranger(Type ~ ., data = Glass, num.trees = 10^3, keep.inbag = TRUE)

oob_ci(rf)

     2.5%     97.5%
0.1401869 0.2523364