Machine Learning - Model Evaluation

Checklist

• Train/validation/test split defined; no peeking.
• Cross-validation for model/feature selection; test used once.
• Metrics match business cost; consider calibration.
• Document dataset versions and seeds; track experiments.

Confusion matrix, ROC/PR (snippet)

from sklearn.metrics import confusion_matrix, roc_auc_score, precision_recall_curve
cm = confusion_matrix(yte, preds)
auc = roc_auc_score(yte, proba)
pr, rc, th = precision_recall_curve(yte, proba)