What binary classification metrics do you know? The pros and cons of each.

Main metrics of binary classification:

1. Accuracy

   • The proportion of correctly classified objects.
   • Plus: simple and intuitive.
   • Minus: performs poorly with imbalanced classes.

2. Precision

   • The proportion of true positives among all predicted positives.
   • Plus: important when false positives are critical.
   • Minus: does not account for missed positives.

3. Recall

   • The proportion of true positives found by the model out of all actual positives.
   • Plus: important when missing positives is critical.
   • Minus: can be high with many false positives.

4. F1-score

   • Harmonic mean of Precision and Recall.
   • Plus: balances between accuracy and completeness.
   • Minus: does not consider true negatives.

5. ROC AUC

   • Measures the model's ability to distinguish between classes at different thresholds.
   • Plus: threshold-independent.
   • Minus: may be less informative with strong imbalance.

6. Confusion Matrix

   • Shows TP, FP, TN, FN for detailed analysis.

The choice of metric depends on the task and which errors are more critical — misses or false alarms.