Probability and Modeling DY0-001 Practice Questions (Page 5)

Questions 21–25

1. 21foundation · easy

   What is the primary purpose of oversampling in data modeling?

   • ATo balance an imbalanced dataset by increasing the number of minority class samples
   • BTo increase the number of features in the dataset
   • CTo remove outliers from the dataset
   • DTo reduce the size of the dataset to improve training speed
   Reveal Solution Discussion Select an answer first
2. 22application · medium

   A healthcare analytics team is building a model to predict patient readmission. The dataset has a binary outcome where only 5% of patients are readmitted. The team notices that the 'insurance type' field is missing for 40% of the readmitted patients but only 10% of the non-readmitted patients. The team decides to drop all rows with missing insurance type before training. What is the most significant risk of this approach?

   • ACausing the model to overfit the training data
   • BIncreasing the variance of the model predictions
   • CIntroducing selection bias that disproportionately removes readmitted patients
   • DCreating multicollinearity among the remaining features
   Reveal Solution Discussion Select an answer first
4. 23expert · hard

   A machine learning engineer is building a binary classifier for a rare disease where only 2% of the population is positive. The engineer has a dataset of 100,000 records. To address the class imbalance, the engineer is considering: (A) random oversampling of the positive class, or (B) SMOTE (Synthetic Minority Over-sampling Technique). The engineer is concerned about overfitting. Which approach is more likely to mitigate overfitting, and why?

   • ABoth approaches have the same risk of overfitting
   • BNeither approach addresses overfitting; only regularization can help
   • CSMOTE, because it creates synthetic samples that are interpolations of existing positive samples, reducing the risk of overfitting compared to simple duplication
   • DRandom oversampling, because it uses real data points, which are more reliable than synthetic ones
   Reveal Solution Discussion Select an answer first
5. 24foundation · easy

   Why is stratification commonly used in train-test splits for classification problems?

   • ATo ensure the test set has no missing values
   • BTo maintain the same class distribution in both training and test sets
   • CTo increase the size of the training set
   • DTo reduce the number of features
   Reveal Solution Discussion Select an answer first
6. 25application · medium

   A risk analyst is comparing two investment portfolios. Portfolio A has a distribution of returns with a high kurtosis, while Portfolio B has a low kurtosis. Both have the same mean and variance. What is the most important implication of the difference in kurtosis for risk assessment?

   • APortfolio A is guaranteed to have more positive returns than Portfolio B
   • BThe difference in kurtosis has no impact on risk because the means are equal
   • CPortfolio B has a higher probability of extreme returns than Portfolio A
   • DPortfolio A has a higher probability of extreme returns (both gains and losses) than Portfolio B
   Reveal Solution Discussion Select an answer first