Classification & Evaluation
You'll be able to
- Explain a classification decision boundary
- Interpret accuracy and why it can mislead
- Understand overfitting vs underfitting
Classification separates inputs into categories with a decision boundary — a learned surface in feature space. A spam classifier might learn: if a message contains many suspicious patterns, label it spam.
Accuracy alone can mislead. If 95% of emails are not spam, a model that always says 'not spam' is 95% accurate but useless. That is why we use precision, recall, and confusion matrices.
A model that memorises training data instead of learning patterns overfits — it fails on new data. One that is too simple underfits. Managing this tension is core to machine learning.
Challenge
Spot the failure
Describe one scenario where 98% accuracy on training data would still make you suspicious of a deployed model.
Knowledge Check
Classification
Why can high accuracy on an imbalanced dataset be misleading?
Overfitting means a model memorises the training data and performs poorly on new data.
Answer all questions to submit.