Theory test: Measuring models honestly

6 questions. Pass with 5 correct.

1. 01 Precision is defined as:

   $TP/(TP+FN)$$TP/(TP+FP)$$(TP+TN)/n$$FP/(FP+TN)$
2. 02 1% of patients have a disease. A model always predicts "healthy". Its accuracy and recall are:

   99% and 99%99% and 0%1% and 100%50% and 50%
3. 03 An AUC of 0.5 means the classifier:

   Is perfectRanks positives above negatives no better than chanceHas 50% precisionAlways predicts positive
4. 04 You standardise the features using the mean and standard deviation of the entire dataset, then split into train and test. This is:

   Correct practiceLeakage: the test set influenced preprocessingOnly a problem for neural networksRequired for cross-validation
5. 05 Why prefer k-fold cross-validation over a single validation split on a small dataset?

   It trains fasterEvery example is used for validation once, and the estimate is less noisyIt replaces the need for a test set entirely, alwaysIt prevents overfitting during training
6. 06 Lowering a classifier's decision threshold usually:

   Raises precision and lowers recallRaises recall and lowers precisionRaises bothChanges neither