Exercise

Plot an ROC curve

As you saw in the video, an ROC curve is a really useful shortcut for summarizing the performance of a classifier over all possible thresholds. This saves you a lot of tedious work computing class predictions for many different thresholds and examining the confusion matrix for each.

My favorite package for computing ROC curves is caTools, which contains a function called colAUC(). This function is very user-friendly and can actually calculate ROC curves for multiple predictors at once. In this case, you only need to calculate the ROC curve for one predictor, e.g.:

colAUC(predicted_probabilities, actual, plotROC = TRUE)

The function will return a score called AUC (more on that later) and the plotROC = TRUE argument will return the plot of the ROC curve for visual inspection.

Instructions

100 XP

model, test, and train from the last exercise using the sonar data are loaded in your workspace.

  • Predict probabilities (i.e. type = "response") on the test set, then store the result as p.
  • Make an ROC curve using the predicted test set probabilities.