Class Imbalance and Resampling Quiz

Answers and explanations appear as you take the quiz.

1. Why is plain accuracy a poor metric for a highly imbalanced classification problem?
   • A model can score very high just by always predicting the majority class
   • Accuracy cannot be computed when classes are imbalanced
   • Accuracy only works for regression
   • It requires perfectly balanced data to be defined
2. What does SMOTE do?
   • Deletes majority-class samples at random
   • Generates synthetic minority samples by interpolating between neighbors
   • Increases the learning rate for rare classes
   • Converts a classification problem into regression
3. Which approach handles imbalance WITHOUT changing the number of training examples?
   • Random oversampling
   • Undersampling the majority
   • Class weighting in the loss function
   • SMOTE
4. What is a key risk of applying oversampling before splitting your data into train and test sets?
   • It makes training slower
   • It deletes the majority class
   • It always causes underfitting
   • Duplicated or synthetic minority points can leak into the test set, inflating scores
5. What is the main downside of random undersampling?
   • It discards potentially useful majority-class data
   • It creates fake data that may be unrealistic
   • It only works for image data
   • It always increases training time
6. Which metric is most appropriate for evaluating a model on rare-positive data?
   • Overall accuracy
   • Precision-Recall AUC or F1 score
   • Mean squared error
   • Number of training epochs
7. What does adjusting the decision threshold after training change?
   • The number of features in the model
   • The class proportions in the training data
   • The trade-off between which examples are labeled positive and negative without changing the training examples
   • The model’s learned weights automatically
8. Which metric does the guide name as an additional option for rare-positive classification?
   • Mean squared error only
   • BLEU score
   • Silhouette coefficient
   • Matthews Correlation Coefficient (MCC)