In Armadilloa16/roc: Receiver Operating Characteristic (ROC)

knitr::opts_chunk$set(collapse = TRUE, comment = "#>")
library(roc)

This document should provide a brief overview of how to generate and plot receiver operating characteristic (ROC) curves and cost functions using the roc package. To do this I'll use a subset of Fisher's (Anderson's) iris data to illustrate:

y       = subset(iris, Species != "virginica")$Sepal.Width
classes = subset(iris, Species != "virginica")$Species == "setosa"

Note that y could be any numeric predictor for any boolean classes. Often, y will be the output of calling predict on a fitted logistic model, for example. The example code below should be pretty self-explanatory.

ROC Curve

roc.df = roc(y, classes)
plot(roc.df$FPR, roc.df$TPR, type="l", xlab="FPR", ylab="TPR")

Cost Function

roc.df$cost.1 = cost(roc.df$FPR, roc.df$TPR)
roc.df$cost.2 = cost(roc.df$FPR, roc.df$TPR, 3)
plot(roc.df$t, roc.df$cost.1, type="l")
plot(roc.df$t, roc.df$cost.2, type="l")

Armadilloa16/roc documentation built on May 5, 2019, 7:06 a.m.