Unit 8: Supervised Learning - Practice Quiz

1 What is the main goal of a classification algorithm?

2 Which type of variable does a classification model usually predict?

3 Which task is an example of binary classification?

4 In supervised classification, what does the training dataset contain?

5 What is multiclass classification?

6 What does the letter represent in KNN?

7 How does KNN commonly assign a class to a new observation?

8 Which measure is commonly used by KNN to identify nearby observations?

9 Why is feature scaling often useful before applying KNN?

10 What does KNN need to retain when making predictions?

11 Naïve Bayes is based on which mathematical theorem?

12 What simplifying assumption does Naïve Bayes make about features?

13 Which application commonly uses Naïve Bayes?

14 How does Naïve Bayes usually select a predicted class?

15 What is a prior probability in Naïve Bayes?

16 What is the main purpose of cross-validation?

17 In -fold cross-validation, how many subsets is the dataset divided into?

18 What does classification accuracy measure?

19 What information does a confusion matrix summarize?

20 What does precision measure in binary classification?

21 A model predicts the probability that a customer will cancel a subscription. If the classification threshold is increased from to , what is the most likely effect?

22 Which task is most appropriately treated as a multiclass classification problem?

23 A classifier performs very well on its training data but poorly on unseen test data. What is the most likely explanation?

24 A binary classifier outputs for an observation. Using a threshold of , which class is assigned?

25 A KNN classifier uses Euclidean distance on age, measured from to , and annual income, measured from to . What should be done before training?

26 For a binary KNN classifier, the five nearest neighbors have labels . What class is predicted using unweighted voting?

27 What is the usual effect of increasing from to a much larger value in a KNN classifier?

28 A KNN model gives too much influence to distant neighbors within the selected neighborhood. Which modification best addresses this issue?

29 Why does KNN often perform poorly when a dataset contains many irrelevant features?

30 The term "naive" in Naive Bayes refers to which modeling assumption?

31 A spam filter has , , and . Before normalization, which value is proportional to ?

32 In a multinomial Naive Bayes text classifier, a word appears in a test document but never appeared in training documents of one class. What prevents the class probability from becoming zero?

33 Which Naive Bayes variant is generally most suitable when features are continuous measurements such as height and weight?

34 Two classes have identical feature likelihoods for a particular observation, but class has a larger prior probability than class . Which prediction will Naive Bayes make?

35 In -fold cross-validation, how many times is each observation normally used for validation?

36 A medical screening classifier should avoid missing patients who actually have a condition. Which metric should receive the most attention?

37 A classifier produces true positives and false positives. What is its precision?

38 A dataset contains negative cases and positive cases. A model predicts every case as negative. Why is its accuracy misleading?

39 Why is stratified cross-validation preferred for an imbalanced classification dataset?

40 A classifier has precision and recall . What is its score?

41 A binary classifier outputs a calibrated posterior probability . A false positive costs 2 units, a false negative costs 8 units, and correct predictions cost 0. Which decision rule minimizes conditional expected cost?

42 A classifier trained with positive-class prior outputs . At deployment, only the class prior changes to ; the class-conditional feature distributions remain unchanged. What is the adjusted posterior probability?

43 A classifier's real-valued score is transformed by a strictly increasing nonlinear function. No ties are introduced. Which statement is necessarily true?

44 A multiclass classifier provides calibrated posteriors. A correct prediction costs 0, an incorrect prediction costs 1, and abstaining costs . Let be the largest class posterior. When is abstention strictly optimal?

45 Three one-vs-one classifiers produce , , and for the same observation. What follows from these pairwise outcomes alone?

46 A KNN model uses Euclidean distance with features age in years and annual income in dollars. Income has a far larger numeric range, although both features are equally relevant. What is the most defensible preprocessing step?

47 Assume every pairwise feature distance is distinct. Which quantity exactly equals the leave-one-out cross-validation error of a 1-NN classifier on a training set?

48 A distance-weighted 3-NN classifier uses weights . The neighbors are class A at distances and , and class B at distance . Which class is predicted?

49 Under standard regularity conditions, which asymptotic choice of supports consistency of KNN as the sample size grows?

50 Two observations have independent irrelevant binary coordinates, each distributed as Bernoulli. For squared Euclidean distance over these coordinates, what is the coefficient of variation?

51 In multinomial Naïve Bayes, the prior odds for classes and are . A token occurs three times and has probabilities under and under . All other token contributions cancel. What are the posterior odds?

52 A Bernoulli Naïve Bayes model has equal class priors. For an observation, and . The probabilities are in class and in class . What is predicted?

53 A one-feature Gaussian Naïve Bayes model has equal priors and class means of 0. Class A has variance 1, and class B has variance 4. For , what does the model predict?

54 In multinomial Naïve Bayes, a class contains 20 total token occurrences over a vocabulary of size 100. A particular token is absent from that class. With additive smoothing , what probability is assigned to the token?

55 An informative binary feature is duplicated exactly and both copies are supplied to Naïve Bayes. Which effect is most accurate?

56 A researcher compares 50 hyperparameter settings and reports the best mean score from the same 10-fold cross-validation used for selection. Which procedure best estimates generalization performance without selection bias?

57 A dataset contains multiple measurements from each patient, and the goal is to predict outcomes for entirely unseen patients. Which cross-validation design is appropriate?

58 A classifier has sensitivity and false-positive rate in a population where prevalence is . Assuming these rates transfer to that population, what is its precision?

59 A classifier has , , , and . What is its Matthews correlation coefficient?

60 A decision threshold is chosen to maximize F1 using the combined out-of-fold predictions from a cross-validation run. The maximum F1 on those same predictions is then reported. What is the main problem?