ROC curve analysis with fbroc

I have written a R-package called fbroc for the fast bootstrap ROC curve analysis. To install package fbroc on your R installation, please use install.packages("fbroc") or go to CRAN. If you want to check out the current development version, go to GitHub.

Start quickly with fbroc

To get familiar with fbroc and to learn what it can do, please use the shiny application.

Current features

  • Very fast bootstrapping of ROC curves.
  • Visualization of confidence regions for the ROC curve.
  • Analysis of the AUC including confidence intervals.
  • Analysis of TPR and FPR.
A ROC curve is shown including the confidence limits.

Confidence region of the ROC curve

High performance

fbroc can calculate 100000 bootstrap iteration for 1000 samples in under 4 seconds. The exact performance depends on your PC.

y <- rep(c(TRUE, FALSE), each = 500)
x <- rnorm(1000) + y
system.time(boot.roc(x, y, n.boot = 100000))
  User      System     elapsed 
  3.59        0.05        3.64 

Planned features

  • Analysis of paired ROC curves to compare two classifiers.
  • Help with finding a cutoff optimized for a specific application.

Instructions for the Shiny application

You can get started right away, using the example data set included with fbroc.

Import your own data
  1. Use a tab-delimited text file as input (Excel can save tables in this format).
  2. The text file must include a column with numerical values. Higher values are assumed to be associated with the positive class.
  3. It must also include a column with the classes. Denote the positive class with TRUE and the negative class with FALSE.

Leave a Reply

Your email address will not be published. Required fields are marked *