Question 1 of 25

What makes learning supervised?

Question 2 of 25

Which task is unsupervised?

Question 3 of 25

What is the main difference between regression and classification?

Question 4 of 25

Which problem is classification?

Question 5 of 25

Why does overfitting happen?

Question 6 of 25

Training accuracy is 99% but test accuracy is 52%. What fits best?

Question 7 of 25

Why do we hold out a test set and not tune the model on it?

Question 8 of 25

Why do we need a validation set (separate from test)?

Question 9 of 25

You preprocess the whole dataset (normalize using global mean/std) before splitting. What went wrong?

Question 10 of 25

When is accuracy a bad primary metric?

Question 11 of 25

In spam filtering, precision answers which question?

Question 12 of 25

In spam filtering, recall answers which question?

Question 13 of 25

A confusion matrix for binary spam detection has high false positives. What is the user-visible problem?

Question 14 of 25

Logistic regression is used for spam detection. Despite the name, it is mainly a ___ model.

Question 15 of 25

Which action reduces overfitting?

Question 16 of 25

K-means clustering is an example of:

Question 17 of 25

Predicting tomorrow’s temperature in °C is best framed as:

Question 18 of 25

You tune hyperparameters using the test set and report final accuracy on the same set. What went wrong?

Question 19 of 25

Underfitting usually means:

Question 20 of 25

When false negatives are costly (e.g. missing cancer on a scan), which metric do you prioritize raising?

Question 21 of 25

Precision answers which question for spam filtering?

Question 22 of 25

Why keep a validation set separate from test?

Question 23 of 25

Which is supervised learning?

Question 24 of 25

Dataset: 98% ham, 2% spam. A model always predicts ham. Accuracy is ~98%. Is this good?

Question 25 of 25

Early stopping during training helps because: