🎯 04 — Classification Evaluation

Classification metrics depend on error cost.

---

Report

from sklearn.metrics import classification_report, confusion_matrix

print(confusion_matrix(y_test, y_pred))
print(classification_report(y_test, y_pred))

---

ROC-AUC and PR-AUC

from sklearn.metrics import roc_auc_score, average_precision_score

roc_auc_score(y_test, y_proba)
average_precision_score(y_test, y_proba)

PR-AUC is often useful for imbalanced positive classes.

---

Threshold Tuning

y_pred = (y_proba >= 0.3).astype(int)

Threshold choice should match business cost.

---

Next

➡️ 05-model-selection-and-tuning