Unit 3: Supervised Machine Learning - Practice Quiz

1 What is the main purpose of linear regression?

2 Which type of relationship does simple linear regression usually model?

3 In linear regression, what is the target variable?

4 What is polynomial regression designed to model?

5 Which expression represents a second-degree polynomial model?

6 What does a decision tree regression model predict?

7 How does a decision tree divide the input data?

8 What is logistic regression commonly used for?

9 Which function is commonly associated with logistic regression?

10 What does a probability near 0.9 usually indicate in binary classification?

11 How does K-Nearest Neighbour classify a new observation?

12 What does the value of represent in K-Nearest Neighbour?

13 Which distance measure is commonly used in K-Nearest Neighbour?

14 What does a decision tree classifier predict?

15 What is a random forest?

16 How does a random forest usually choose a class for classification?

17 Naïve Bayes is based mainly on which mathematical concept?

18 What assumption does the naïve part of Naïve Bayes make?

19 What does high bias usually indicate?

20 What is overfitting?

21 A linear regression model is . What is the predicted value when and ?

22 A linear regression model has a coefficient of for advertising cost, measured in thousands of dollars. How should this coefficient be interpreted when other variables remain fixed?

23 A quadratic regression model is . At which value of does the model reach its maximum predicted value?

24 A degree-10 polynomial model has very low training error but much higher validation error. Which change is most likely to improve its generalization?

25 A regression-tree leaf contains target values , , , and . Assuming standard squared-error training, what prediction is made for a new sample reaching this leaf?

26 Why can the predictions of an unpruned decision tree regressor change sharply when an input crosses a split threshold?

27 A logistic regression model outputs a probability of for the positive class. What class is predicted using a threshold of ?

28 In a logistic regression model, a feature has coefficient . Holding other features constant, what happens to the odds of the positive class when this feature increases by one unit?

29 A K-Nearest Neighbour classifier uses age values from to and annual income values from to . What should usually be done before computing Euclidean distances?

30 A 5-NN classifier finds neighbor labels for a new sample. What class will it predict using majority voting?

31 Increasing from to a moderately larger value in K-Nearest Neighbour classification generally has which effect?

32 A decision tree classifier is considering two splits. Split P produces mostly pure child nodes, while Split Q leaves both classes heavily mixed. Which split is normally preferred?

33 A decision tree classifier perfectly classifies its training set but performs poorly on unseen data. Which adjustment most directly addresses this problem?

34 Why does a random forest usually generalize better than a single deep decision tree?

35 In a random forest, what is the main purpose of considering only a random subset of features at each split?

36 For a class , a Naïve Bayes model has , , and . What is the unnormalized posterior score for ?

37 A Gaussian Naïve Bayes classifier is most directly suited to which type of input?

38 A model has high error on both its training set and validation set, and the two errors are similar. What is the most likely diagnosis?

39 A model's training accuracy is , but its validation accuracy is . Which observation best describes this result?

40 Validation error decreases as model complexity grows, reaches a minimum, and then begins increasing while training error continues decreasing. Which model should generally be selected?

41 A linear regression model is fitted using two standardized predictors, and . Their correlation is , and both predictors have substantial marginal correlation with the target. Which outcome is most likely when ordinary least squares is used?

42 For a regression model with an intercept, the residuals are orthogonal to every included predictor. If the model predicts the training observations exactly, which statement must hold?

43 A single training observation has an unusually large predictor value and lies close to the fitted regression line. Another observation has an ordinary predictor value but a very large vertical residual. Which diagnosis is most appropriate?

44 A degree-5 polynomial fits a small training set nearly perfectly but produces extreme oscillations just outside the observed range. Which intervention most directly addresses the primary failure?

45 Suppose the true relationship is , but a degree-1 model is fitted over a narrow interval centered at . What is the main consequence of omitting the quadratic term?

46 A decision tree regressor considers a split that creates two leaves. The left leaf contains values , and the right leaf contains . What prediction does the tree make for a new point assigned to the right leaf, assuming squared-error loss?

47 A regression tree is expanded until every training sample occupies its own leaf. Training mean squared error becomes zero. Which statement best describes the expected generalization behavior?

48 In logistic regression, a coefficient for feature is . Holding all other variables fixed, which change in the odds of the positive class results from increasing by one unit?

49 A classifier outputs well-calibrated probabilities, but the positive class is rare and the cost of false negatives is much higher than the cost of false positives. What is the most defensible decision rule?

50 A K-nearest neighbour classifier uses Euclidean distance on two features: annual income measured in dollars and age measured in years. Which preprocessing is most important before fitting the classifier?

51 As increases substantially in a K-nearest neighbour classifier, which bias-variance change is generally expected, assuming the distance metric remains fixed?

52 For a binary classification node containing equal numbers of both classes, a candidate split creates two pure child nodes of equal size. What is the weighted Gini impurity after the split?

53 Why does randomly selecting a subset of features at each split typically improve a random forest compared with averaging fully correlated decision trees?

54 A Naïve Bayes classifier assigns class to an instance using . One feature value has zero estimated likelihood in every observed training instance of class . What is the standard remedy?

55 Naïve Bayes assumes conditional independence of features given the class. If two highly correlated features are both included, which issue can arise even when classification accuracy remains reasonable?

56 A model has nearly identical training and validation errors, and both errors are large relative to a more flexible benchmark. Which diagnosis is most consistent with this evidence?

57 Two models have the same average validation error across many splits. Model A's error varies widely between splits, while Model B's error is stable. Which model has greater estimation variance?

58 A polynomial model's training error decreases monotonically as degree increases, while validation error decreases initially and then rises. At the degree where validation error is smallest, what is the best interpretation?

59 A preprocessing step computes the mean and standard deviation using the entire dataset before cross-validation. The step does not use target labels. Why can this still lead to optimistic validation estimates?

60 A model performs extremely well on randomly shuffled test data but poorly when evaluated on a later time period. Which explanation is most plausible?