Unit 5: Model Evaluation and Improvement - Practice Quiz

CSE252 — Introduction To Artificial Intelligence And Machine Learning 60 Questions
0 Correct 0 Wrong 60 Left
0/60

1 What does Mean Absolute Error (MAE) measure?

Regression Metrics (MAE, MSE, RMSE, R² Score) Easy
A. The average absolute difference between predictions and actual values
B. The proportion of positive predictions
C. The average squared difference between predictions and actual values
D. The percentage of correctly classified samples

2 Which regression metric gives greater importance to larger errors by squaring them?

Regression Metrics (MAE, MSE, RMSE, R² Score) Easy
A. Accuracy
B. Recall
C. MAE
D. MSE

3 What is the relationship between RMSE and MSE?

Regression Metrics (MAE, MSE, RMSE, R² Score) Easy
A. RMSE is the square of MSE
B. RMSE is the average of MSE values
C. RMSE is the square root of MSE
D. RMSE is unrelated to MSE

4 What does an score of 1 generally indicate?

Regression Metrics (MAE, MSE, RMSE, R² Score) Easy
A. The model explains none of the variation
B. The model perfectly explains the observed variation
C. The model makes completely random predictions
D. The model has exactly one prediction error

5 For MAE, MSE, and RMSE, which value usually indicates better regression performance?

Regression Metrics (MAE, MSE, RMSE, R² Score) Easy
A. A value closer to one
B. A value above one hundred
C. A value equal to the number of features
D. A value closer to zero

6 What does a confusion matrix summarize?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Easy
A. Correct and incorrect classification predictions
B. The relationship between two continuous variables
C. The average value of a regression target
D. The number of model training epochs

7 What does classification accuracy measure?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Easy
A. The fraction of predicted positives that are correct
B. The fraction of actual positives that are found
C. The fraction of all predictions that are correct
D. The average distance between predicted values

8 Precision is mainly concerned with which question?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Easy
A. Of the actual positives, how many are identified?
B. Of all samples, how many predictions are correct?
C. Of the predicted positives, how many are actually positive?
D. How large are the numerical prediction errors?

9 What does recall measure?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Easy
A. The average squared prediction error
B. The proportion of actual positives correctly identified
C. The proportion of all predictions that are correct
D. The proportion of predicted positives that are correct

10 The F1-score combines which two metrics?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Easy
A. Accuracy and ROC-AUC
B. Specificity and RMSE
C. Precision and recall
D. MAE and MSE

11 What does ROC-AUC generally evaluate?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Easy
A. The number of features used by a model
B. The time needed to train a model
C. The average size of regression errors
D. A model's ability to distinguish between classes

12 In a binary classification problem, what is a false positive?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Easy
A. A negative case predicted as positive
B. A negative case predicted as negative
C. A positive case predicted as negative
D. A positive case predicted as positive

13 What is the main purpose of cross-validation?

Model Improvement (Cross Validation, Hyperparameter Tuning, Grid Search, Random Search) Easy
A. To guarantee perfect predictions
B. To increase the number of target labels
C. To remove every feature from the dataset
D. To estimate how well a model generalizes

14 In -fold cross-validation, what happens to the dataset?

Model Improvement (Cross Validation, Hyperparameter Tuning, Grid Search, Random Search) Easy
A. It is reduced to one training example
B. It is divided into parts
C. It is sorted only by prediction value
D. It is converted into target variables

15 What is a hyperparameter?

Model Improvement (Cross Validation, Hyperparameter Tuning, Grid Search, Random Search) Easy
A. A setting chosen before model training
B. A label created by the confusion matrix
C. A prediction produced after model training
D. An error calculated from the test set

16 What does hyperparameter tuning try to find?

Model Improvement (Cross Validation, Hyperparameter Tuning, Grid Search, Random Search) Easy
A. The exact value of every missing feature
B. A new classification label for each sample
C. The names of all training examples
D. A useful combination of model settings

17 How does grid search usually select hyperparameter values?

Model Improvement (Cross Validation, Hyperparameter Tuning, Grid Search, Random Search) Easy
A. It tests only the first combination
B. It tests all specified combinations
C. It chooses values without evaluating the model
D. It changes the target labels during training

18 How does random search choose hyperparameter combinations?

Model Improvement (Cross Validation, Hyperparameter Tuning, Grid Search, Random Search) Easy
A. By selecting only the smallest parameter values
B. By sampling combinations from specified ranges
C. By using the test labels to set each parameter
D. By testing every possible combination in order

19 What is the main goal of Explainable AI (XAI)?

Introduction to Explainable AI Easy
A. To ensure that every model has zero prediction errors
B. To make models use only numerical input features
C. To make model decisions easier for people to understand
D. To replace all datasets with manually written rules

20 Why can explainability be important in an AI system?

Introduction to Explainable AI Easy
A. It guarantees that the training data contains no bias
B. It removes the need to evaluate model performance
C. It makes every prediction independent of input data
D. It can help users identify and trust appropriate model behavior

21 A regression model produces errors of , , , and . What is the model's MAE?

Regression Metrics (MAE, MSE, RMSE, R² Score) Medium
A.
B.
C.
D.

22 Two regression models have error sets. Model A has errors , while Model B has errors . Which model has the lower MAE?

Regression Metrics (MAE, MSE, RMSE, R² Score) Medium
A. Model A, because its MAE is
B. Both models, because their MAEs are equal
C. Model A, because its MAE is
D. Model B, because its MAE is

23 A model's residuals are . What is its MSE?

Regression Metrics (MAE, MSE, RMSE, R² Score) Medium
A.
B.
C.
D.

24 A regression model has an MSE of . What is its RMSE, and what does this value represent?

Regression Metrics (MAE, MSE, RMSE, R² Score) Medium
A. , the total squared error
B. , the typical error in the target's units
C. , the average absolute error
D. , the typical error in squared units

25 A regression model has . Which interpretation is most appropriate?

Regression Metrics (MAE, MSE, RMSE, R² Score) Medium
A. The model correctly predicts of observations
B. The model makes predictions with accuracy
C. The model explains of the target variance
D. The model has an average error of units

26 A classifier produces true positives, false positives, false negatives, and true negatives. What is its precision?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Medium
A.
B.
C.
D.

27 Using true positives, false positives, false negatives, and true negatives, what is the classifier's recall?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Medium
A.
B.
C.
D.

28 A medical screening model has high recall but low precision. What does this most likely mean?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Medium
A. It identifies most positives but raises many false alarms
B. It produces few positive predictions overall
C. It classifies both classes with equal accuracy
D. It misses most positive cases

29 A classifier has precision and recall . What is its F1-score?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Medium
A.
B.
C.
D.

30 A fraud dataset contains fraudulent transactions. A model predicts every transaction as legitimate and achieves accuracy. Which metric would better reveal its failure to detect fraud?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Medium
A. The number of true negatives
B. Recall for the fraud class
C. Accuracy
D. The total number of observations

31 What does an ROC-AUC value of generally indicate for a binary classifier?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Medium
A. A model with no false positives
B. Performance similar to random ranking
C. Perfect classification
D. Correct predictions for half the samples

32 When the cost of missing a positive case is much higher than the cost of a false alarm, which metric should receive greater emphasis?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Medium
A. Recall
B. Precision
C. Specificity
D. Overall accuracy

33 In 5-fold cross-validation, how is the dataset typically used during one complete evaluation cycle?

Model Improvement (Cross Validation, Hyperparameter Tuning, Grid Search, Random Search) Medium
A. It is split into five parts, each used once for validation
B. Each fold is used only for final testing
C. Five separate datasets are collected for testing
D. The model is trained only on one part

34 Why is cross-validation often preferred to a single train-validation split when the dataset is small?

Model Improvement (Cross Validation, Hyperparameter Tuning, Grid Search, Random Search) Medium
A. It guarantees that the model cannot overfit
B. It removes the need for a test set
C. It uses multiple validation partitions for a more stable estimate
D. It always increases the final training accuracy

35 Which of the following is a hyperparameter for a decision tree model?

Model Improvement (Cross Validation, Hyperparameter Tuning, Grid Search, Random Search) Medium
A. A training example's feature value
B. The tree's maximum depth
C. The predicted class label
D. The residual for one observation

36 A grid search tests 3 values for a learning rate and 4 values for maximum tree depth. Ignoring cross-validation repetitions, how many hyperparameter combinations are evaluated?

Model Improvement (Cross Validation, Hyperparameter Tuning, Grid Search, Random Search) Medium
A.
B.
C.
D.

37 Why can random search be more efficient than grid search when only a few hyperparameters strongly affect performance?

Model Improvement (Cross Validation, Hyperparameter Tuning, Grid Search, Random Search) Medium
A. It tests every possible combination
B. It eliminates the need for validation data
C. It samples combinations across the search space
D. It guarantees the global optimum

38 A model performs very well on training folds but poorly on validation folds during cross-validation. Which change is most likely to help?

Model Improvement (Cross Validation, Hyperparameter Tuning, Grid Search, Random Search) Medium
A. Train and evaluate on the same samples
B. Remove the validation folds
C. Increase model complexity substantially
D. Use regularization or reduce model complexity

39 What is the primary goal of Explainable AI?

Introduction to Explainable AI Medium
A. To make model decisions understandable to people
B. To guarantee predictions are unbiased
C. To make every model smaller
D. To replace evaluation metrics

40 A feature-importance analysis shows that income contributes strongly to a loan model's predictions. What does this result mean?

Introduction to Explainable AI Medium
A. Income must be the fairest feature
B. Income causes the approved loans
C. Income alone determines every prediction
D. Income is associated with the model's predictive behavior

41 A regression model has residuals . What are its MAE, MSE, and RMSE, respectively?

Regression Metrics (MAE, MSE, RMSE, R² Score) Hard
A.
B.
C.
D.

42 Two models are evaluated on the same regression targets. Model A has MAE and RMSE , while Model B has MAE and RMSE . Which conclusion is most defensible?

Regression Metrics (MAE, MSE, RMSE, R² Score) Hard
A. Model B is better under absolute-error loss
B. Model A has fewer large errors than Model B
C. The metrics cannot be compared because their units differ
D. Model B has uniformly smaller prediction errors

43 A model obtains on a test set. The target variance is later found to be much smaller in a second test set, while the model's RMSE remains unchanged. What is the most likely effect on ?

Regression Metrics (MAE, MSE, RMSE, R² Score) Hard
A. It becomes zero because the target variance is smaller
B. It remains exactly
C. It increases because RMSE is unchanged
D. It decreases because the residual variance is larger relative to target variance

44 A test set contains one extreme target value caused by a measurement error. Which metric is most likely to change disproportionately if the model's prediction for that point is poor?

Regression Metrics (MAE, MSE, RMSE, R² Score) Hard
A. R², because it ignores residual magnitude
B. The median absolute error, because it emphasizes extremes
C. MSE, because it squares deviations
D. MAE, because it uses absolute deviations

45 A regression model predicts the mean target value for every test example. Under ordinary definitions, which statement is correct when the test target variance is nonzero?

Regression Metrics (MAE, MSE, RMSE, R² Score) Hard
A. Its is undefined because constant predictions are invalid
B. Its is negative and its MAE must equal zero
C. Its is and its RMSE equals the target standard deviation
D. Its is and its RMSE is zero

46 A classifier produces , , , and . Which pair is correct?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Hard
A. Precision , recall
B. Precision , recall
C. Precision , recall
D. Precision , recall

47 In a highly imbalanced fraud-detection problem, a classifier predicts every transaction as legitimate. Which metric can appear excellent while the model remains operationally useless?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Hard
A. Fraud recall
B. Fraud precision
C. Accuracy
D. Fraud F1-score

48 For a binary classifier, lowering the decision threshold generally produces which change, assuming the score ordering stays fixed?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Hard
A. More predicted negatives and higher recall
B. More predicted positives and higher recall
C. No change in the confusion matrix or threshold metrics
D. Fewer predicted positives and higher precision

49 A classifier has precision and recall . What is its F1-score?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Hard
A.
B.
C.
D.

50 A model's ROC-AUC is , but its precision at the deployment threshold is only $0.12. Which explanation is most plausible?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Hard
A. The model must have an invalid ROC curve
B. ROC-AUC directly equals precision averaged over thresholds
C. The model ranks examples well, but positives are rare and the threshold is poorly selected
D. High ROC-AUC guarantees high precision at every threshold

51 Two classifiers have identical ROC-AUC values. Classifier A has higher precision at the threshold required by the application. What can be concluded?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Hard
A. Classifier B has higher recall at every threshold
B. Classifier A must have higher accuracy on every dataset
C. Classifier A is preferable for that operating requirement
D. Classifier B has a strictly better ranking function

52 A screening system must minimize missed positive cases, and false positives can be investigated manually. Which evaluation priority is most appropriate?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Hard
A. Maximize recall, then monitor precision
B. Maximize precision, then ignore recall
C. Maximize accuracy because costs are symmetric
D. Maximize ROC-AUC at one fixed threshold

53 A multiclass classifier has per-class recalls of , , and . Which averaging strategy gives each class equal importance regardless of its sample count?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Hard
A. Macro-averaged recall
B. Weighted-average recall
C. Majority-class recall
D. Micro-averaged recall

54 A classifier's accuracy rises after deployment because the positive class becomes much rarer, while its positive recall falls sharply. Which interpretation is strongest?

Confusion Matrix and Classification Metrics (Accuracy, Precision, Recall, F1-score, ROC-AUC) Hard
A. Recall must be wrong because accuracy and recall always move together
B. The classifier became perfectly calibrated
C. The model necessarily improved because accuracy increased
D. The class distribution changed, making accuracy misleading for the positive class

55 Why should preprocessing steps such as feature scaling be fitted separately inside each training fold during cross-validation?

Model Improvement (Cross Validation, Hyperparameter Tuning, Grid Search, Random Search) Hard
A. To increase the validation score through additional data leakage
B. To prevent validation-fold information from influencing the fitted transformation
C. To guarantee that all models use identical coefficients
D. To make every fold contain the same number of samples

56 A dataset contains repeated measurements from the same patients. Randomly assigning rows to cross-validation folds causes very high validation scores. Which change best addresses the problem?

Model Improvement (Cross Validation, Hyperparameter Tuning, Grid Search, Random Search) Hard
A. Use the test set during hyperparameter selection
B. Remove regularization so patient patterns are retained
C. Use more random folds with the same row-level split
D. Use group-based cross-validation with patient identity as the group

57 A grid search evaluates five values for each of four independent hyperparameters. How many configurations are evaluated before accounting for cross-validation folds?

Model Improvement (Cross Validation, Hyperparameter Tuning, Grid Search, Random Search) Hard
A. 125
B. 25
C. 20
D. 625

58 When is random search often more efficient than grid search for hyperparameter optimization?

Model Improvement (Cross Validation, Hyperparameter Tuning, Grid Search, Random Search) Hard
A. When every possible configuration must be tested exhaustively
B. When the search space contains no continuous values
C. When only a few hyperparameters strongly influence performance
D. When all hyperparameters affect performance equally

59 A practitioner repeatedly chooses hyperparameters based on cross-validation results and reports the best cross-validation score as final performance. What is the primary methodological problem?

Model Improvement (Cross Validation, Hyperparameter Tuning, Grid Search, Random Search) Hard
A. The training set is always smaller than the test set
B. Hyperparameters are learned from target values by definition
C. Cross-validation cannot compare hyperparameters
D. The validation results have become part of the model-selection process

60 A feature attribution method assigns positive contributions to several correlated features. Why should these individual attributions be interpreted cautiously?

Introduction to Explainable AI Hard
A. Correlated features can share or redistribute credit across equivalent explanations
B. Correlation guarantees that every attribution is causally correct
C. Positive contributions prove that removing each feature improves accuracy
D. Attributions are always independent of the model's training data