Accuracy

• misleading for class imbalance (such as churn rate) → just answer with the biggest classes still get good score
• from the above image
  • even though we predict there’s no one gonna churn → it still get acc = 0.72
  • so from the previous lesson that our model can predict acc~0.8 → might not confirm good performance

Confusion matrix

• true positive → pred positive , actually positive
• true negative → pred negative , actually negative
• false positive → pred positive , actually negative
• false negative → pred negative, actually positive

Recall

• how much target label (1) was correctedly predict
• tp / (tp + fn)

Precision