Model Evaluation

Train/Test Split

• Never evaluate on training data
• Typical: 80% train, 20% test
• Overfitting hides in train metrics

Confusion Matrix

• TN, FP, FN, TP
• Shows error types
• Foundation for other metrics

Precision

• TP / (TP + FP)
• Of predicted positives, how many correct?
• High when FP costly

Recall

• TP / (TP + FN)
• Of actual positives, how many found?
• High when FN costly

F1 Score

• 2 × P × R / (P + R)
• Harmonic mean
• Good for imbalanced data

ROC and AUC

• TPR vs FPR at various thresholds
• AUC: single number summary
• 1.0 = perfect, 0.5 = random

Cross-Validation

• K-fold: train/test K times
• Average scores
• More robust than single split