Mastering Logistic Regression for Binary u0026 Multiclass Classification Quiz

1. Question 1

   Given a dataset with an imbalanced binary target variable, which approach best addresses the tendency of logistic regression to favor the majority class during model training?

   1. A. Oversampling the minority class or using class weighting
   2. B. Increasing the learning rate hyperparameter
   3. C. Applying principal component analysis to input features
   4. D. Reducing the decision boundary threshold below 0.5
   5. E. Performing k-means clustering before fitting the model

2. Question 2

   In multinomial logistic regression applied to a classification task with five classes (A, B, C, D, E), how many sets of coefficients (including intercepts) does the final model estimate, assuming one-vs-rest (OvR) strategy?

   1. A. Five sets, one for each class
   2. B. Four sets, as one class is the reference
   3. C. Ten sets, two per class
   4. D. One single set shared by all classes
   5. E. Six sets corresponding to all pairwise class combinations

3. Question 3

   Which one of the following statements correctly describes the difference in the mathematical formulation between binary logistic regression and multinomial logistic regression?

   1. A. Binary logistic regression uses a single sigmoid function, while multinomial uses multiple softmax functions
   2. B. Binary logistic regression models log-odds for two classes; multinomial models log-odds for each class relative to a baseline
   3. C. Both models use exactly the same loss function and probability output
   4. D. Multinomial logistic regression relies on least squares estimation instead of maximum likelihood
   5. E. There are no differences; both are mathematically identical

4. Question 4

   Consider a logistic regression model predicting whether a transaction is fraudulent (1) or legitimate (0); if the model’s predicted probability for fraud is 0.95 but the actual event is legitimate, which metric would reflect the impact of this confident misclassification most strongly?

   1. A. Log loss (cross-entropy loss)
   2. B. Precision
   3. C. Recall
   4. D. F1 score
   5. E. Euclidian loss

5. Question 5

   When interpreting the coefficient of a feature in binary logistic regression, what does exponentiating the coefficient (e.g., exp(β₁)) represent?

   1. A. The change in probability for a one-unit increase in the predictor
   2. B. The odds ratio representing the multiplicative change in odds for a one-unit increase in the predictor
   3. C. The slope of the linear regression line for that feature
   4. D. The regularized inverse-odds of the outcome
   5. E. The p-value of the hypothesis test for that feature