5.2 Evaluating Classification Models

• Accuracy: \(\frac{TP+TN}{total}\)
  • yardstick::accuracy()
• Confusion matrix: columns = predictions, rows = actual, diagonal = correctly classified
  • yardstick::conf_mat()
• Recall (aka Sensitivity): \(\frac{TP}{TP+FN}\)
  • “Of the true things, how many did the model ‘remember’?”
  • yardstick::recall()
• Specificity: \(\frac{TN}{TN+FP}\)
  • “How good is the model at picking out bad things?”
  • yardstick::spec() or yardstick::specificity()
• Precision: \(\frac{TP}{TP+FP}\)
  • “What portion of the predicted true things are true?”
  • yardstick::precision()