Data Science Model Performance and Evaluation 4

Question 1

Data leakage during evaluation typically causes:

Accepted Answer

Overly optimistic performance that fails in production

Answer

Pessimistic performance estimates

Answer

No effect on metrics

Answer

Lower training accuracy

Question 2

When evaluating a model on time-series data, which validation strategy is appropriate?

Accepted Answer

Time-based (forward-chaining) splits

Answer

Random k-fold shuffling

Answer

Stratified shuffle split

Answer

Leave-one-out on shuffled rows

Question 3

For a multi-class classification problem, macro-averaged F1 differs from micro-averaged F1 because macro:

Accepted Answer

Treats all classes equally regardless of size

Answer

Weights classes by their frequency

Answer

Ignores minority classes

Answer

Is only for binary problems

Question 4

Adjusting the classification threshold from 0.5 to 0.7 will most likely:

Accepted Answer

Increase precision, decrease recall

Answer

Increase recall, decrease precision

Answer

Increase both equally

Answer

Have no effect on either

Question 5

Which statement about adjusted R-squared is correct?

Accepted Answer

It penalizes adding predictors that don't improve the model

Answer

It always increases when adding predictors

Answer

It is identical to R-squared

Answer

It only applies to classification

Question 6

Log loss (cross-entropy) heavily penalizes a model when it:

Accepted Answer

Is confidently wrong (high probability on the wrong class)

Answer

Predicts the correct class with high confidence

Answer

Abstains from predicting

Answer

Predicts probabilities near 0.5

Question 7

Matthews Correlation Coefficient (MCC) is valued for binary classification because it:

Accepted Answer

Gives a balanced measure even with very imbalanced classes

Answer

Ignores the negative class entirely

Answer

Only works when classes are equal

Answer

Is identical to accuracy