A classifier predicts a category rather than a continuous number. It answers questions such as whether an email is spam or whether an image shows a dog, cat, or bird. section 8.0
Logistic regression is a common classifier: it first predicts the probability of a label, then converts that probability into a class using a threshold such as 0.50. section 8.0 section 8.11
Evaluate a classifier with care:
- Do not rely on accuracy alone when classes are imbalanced; a model can appear accurate by favoring the overrepresented class. section 8.11
- Use precision and recall to examine the trade-off between false alarms and missed cases. section 8.11
- Hold back unseen data with a train/test split or cross-validation, and compare performance across thresholds with ROC or precision-recall curves. section 8.11
