6.22 Measuring classifier performance

• Some more measure of model accuracy/classifier performance
• Correct Classification Rate (CCR) [as opposed to Error Rate]
  • For what proportion of the units does the correct output $y_{i}$ match the classifier (predicted) output $\hat{y}_{i}$?
  • Proportion of reoffenders correctly predicted
• False Positive Rate (FPR)
  • For what proportion of the units for which the correct output $y_{i}$ is negative is the classifier output $\hat{y}_{i}$ positive?
  • Proportion of NON reoffenders (NON recidivators) predicted as reoffenders (recidivators)
• False Negative Rage (FNR): Opposite of above