ROC Curves & AUC

Your spam classifier reports 92% accuracy. Sounds great — until you realize 85% of your data is ham. A model that always predicts "ham" would score 85%. Accuracy is lying to you. The ROC curve tells the whole story: it plots your model's true positive rate against its false positive rate at every possible threshold. The area under that curve (AUC) gives you one number that captures real performance. Build it and see.