auc: Area under the ROC Curve

In egenn/rtemis: Machine Learning and Visualization

Area under the ROC Curve

Description

Get the Area under the ROC curve to assess classifier performance.

Usage

auc(
  preds,
  labels,
  method = c("pROC", "ROCR", "auc_pairs"),
  verbose = FALSE,
  trace = 0
)

Arguments

preds	Numeric, Vector: Probabilities or model scores (e.g. c(.32, .75, .63), etc)
labels	True labels of outcomes (e.g. c(0, 1, 1))
method	Character: "pROC", "auc_pairs", or "ROCR": Method to use. Will use pROC::roc, auc_pairs, ROCR::performance, respectively.
verbose	Logical: If TRUE, print messages to output
trace	Integer: If > 0, print more messages to output

Details

Important Note: We assume that true labels are a factor where the first level is the "positive" case, a.k.a. the event. All methods used here, "pROC", "auc_pairs", "ROCR", have been setup to expect this. This goes against the default setting for both "pROC" and "ROCR", which will not give an AUC less than .5 because they will reorder levels. We don't want this because you can have a classifier perform worse than .5 and it can be very confusing if levels are reordered automatically and different functions give you different AUC.

Author(s)

EDG

Examples

## Not run: 
preds <- c(0.7, 0.55, 0.45, 0.25, 0.6, 0.7, 0.2)
labels <- factor(c("a", "a", "a", "b", "b", "b", "b"))
auc(preds, labels, method = "ROCR")
auc(preds, labels, method = "pROC")
auc(preds, labels, method = "auc_pairs")

## End(Not run)

egenn/rtemis documentation built on March 28, 2024, 12:53 p.m.