Back to Subject

Unit 3: Machine Learning - Practice Quiz

INT428 — Artificial Intelligence Essentials 60 Questions
0 Correct 0 Wrong 60 Left
0/60

1 What is the probability of getting heads when flipping a single fair coin?

Probability Easy
A.
B.
C.
D.

2 What is the range of any valid probability value?

Probability Easy
A. Between and
B. Between and
C. Any real number
D. Between and

3 Two events are called independent when:

Probability Easy
A. Their probabilities sum to
B. One event always prevents the other
C. The occurrence of one does not affect the other
D. They always happen together

4 Which measure represents the average of a set of numbers?

Statistics Easy
A. Range
B. Median
C. Mean
D. Mode

5 The median of a dataset is the value that:

Statistics Easy
A. Lies in the middle when data is ordered
B. Occurs most frequently
C. Is the largest value
D. Equals the sum of all values

6 What does standard deviation measure in a dataset?

Statistics Easy
A. The middle value of the data
B. The most common value
C. The total count of values
D. The spread of data around the mean

7 In machine learning, a single row of features is often represented as a:

Linear algebra (applied focus) Easy
A. Vector
B. Determinant
C. Matrix inverse
D. Scalar

8 A dataset with rows as samples and columns as features is best represented as a:

Linear algebra (applied focus) Easy
A. Single number
B. Matrix
C. Boolean value
D. Scalar

9 What is the result of the dot product of two vectors?

Linear algebra (applied focus) Easy
A. A scalar
B. A vector
C. A matrix
D. A determinant

10 Which type of learning uses labeled data for training?

Supervised, unsupervised, and reinforcement learning: concepts and real-world use Easy
A. Supervised learning
B. Unsupervised learning
C. Clustering
D. Reinforcement learning

11 Grouping customers into segments without predefined labels is an example of:

Supervised, unsupervised, and reinforcement learning: concepts and real-world use Easy
A. Unsupervised learning
B. Reinforcement learning
C. Regression
D. Supervised learning

12 In reinforcement learning, an agent learns by:

Supervised, unsupervised, and reinforcement learning: concepts and real-world use Easy
A. Grouping similar data points
B. Copying labeled examples
C. Receiving rewards and penalties
D. Reducing the number of features

13 Predicting house prices from features like size and location is an example of a:

Supervised, unsupervised, and reinforcement learning: concepts and real-world use Easy
A. Regression task
B. Classification task
C. Clustering task
D. Reward task

14 What is the main goal of feature engineering?

Feature engineering and model evaluation (cross-validation, precision, recall) Easy
A. To create useful input features for a model
B. To reward the model for correct answers
C. To label the training data
D. To increase the number of data samples

15 Precision is defined as the ratio of:

Feature engineering and model evaluation (cross-validation, precision, recall) Easy
A. True positives to all actual positives
B. True positives to all predicted positives
C. False positives to true negatives
D. Correct predictions to total predictions

16 Recall measures the proportion of:

Feature engineering and model evaluation (cross-validation, precision, recall) Easy
A. Predicted positives that are correct
B. Actual positives correctly identified
C. Negatives correctly rejected
D. Total correct predictions overall

17 What is the purpose of cross-validation?

Feature engineering and model evaluation (cross-validation, precision, recall) Easy
A. To reduce the number of features
B. To assess how well a model generalizes
C. To label the raw dataset
D. To increase the training set size

18 Bayes' theorem is used to compute which type of probability?

Bayes theorem, Bayesian networks, and probabilistic reasoning Easy
A. Posterior (conditional) probability
B. Marginal probability only
C. Total probability only
D. Joint probability only

19 In Bayes' theorem , the term is called the:

Bayes theorem, Bayesian networks, and probabilistic reasoning Easy
A. Likelihood
B. Posterior
C. Prior
D. Evidence

20 A Bayesian network is best described as a:

Bayes theorem, Bayesian networks, and probabilistic reasoning Easy
A. Sequence of reward signals
B. Table of raw feature values
C. List of unlabeled data points
D. Graph showing probabilistic relationships between variables

21 A bag contains 3 red and 2 blue balls. Two balls are drawn without replacement. What is the probability that both are red?

Probability Medium
A.
B.
C.
D.

22 If events and are independent with and , what is ?

Probability Medium
A.
B.
C.
D.

23 A dataset has values . What is the mean?

Statistics Medium
A.
B.
C.
D.

24 In a right-skewed distribution, which ordering of central tendency measures is typically correct?

Statistics Medium
A. Mode > Median > Mean
B. Mean = Median = Mode
C. Median > Mean > Mode
D. Mean > Median > Mode

25 Given vectors and , what is the dot product ?

Linear algebra (applied focus) Medium
A.
B.
C.
D.

26 In machine learning, why is matrix multiplication central to computing outputs of a fully connected neural network layer?

Linear algebra (applied focus) Medium
A. It removes correlated features
B. It normalizes the data to zero mean
C. It sorts the input features
D. It applies the learned weights to every input feature simultaneously as a linear transformation before activation

27 What does the identity matrix satisfy for any compatible matrix ?

Linear algebra (applied focus) Medium
A.
B.
C.
D.

28 A company wants to group customers into segments based on purchasing behavior without any predefined labels. Which learning paradigm fits best?

Supervised, unsupervised, and reinforcement learning: concepts and real-world use Medium
A. Supervised learning
B. Unsupervised learning
C. Semi-supervised regression
D. Reinforcement learning

29 Which scenario is the clearest example of reinforcement learning?

Supervised, unsupervised, and reinforcement learning: concepts and real-world use Medium
A. An agent learns to play a game by receiving rewards and penalties for its moves over time
B. A regression model estimates temperature
C. An algorithm clusters news articles by topic
D. A model predicts house prices from labeled data

30 Predicting whether an email is spam or not spam using a dataset of labeled emails is an example of which task?

Supervised, unsupervised, and reinforcement learning: concepts and real-world use Medium
A. Reinforcement learning
B. Unsupervised clustering
C. Supervised classification
D. Dimensionality reduction

31 A classifier produces 40 true positives, 10 false positives, and 5 false negatives. What is its precision?

Feature engineering and model evaluation (cross-validation, precision, recall) Medium
A.
B.
C.
D.

32 For the same classifier with 40 true positives, 10 false positives, and 5 false negatives, what is its recall?

Feature engineering and model evaluation (cross-validation, precision, recall) Medium
A.
B.
C.
D.

33 What is the main advantage of -fold cross-validation over a single train/test split?

Feature engineering and model evaluation (cross-validation, precision, recall) Medium
A. It provides a more reliable estimate of model performance by using every data point for both training and validation across folds
B. It always reduces training time
C. It guarantees higher accuracy
D. It eliminates the need for a test set

34 Which feature engineering technique is most appropriate for a categorical variable like 'color' with values {red, green, blue} for a linear model?

Feature engineering and model evaluation (cross-validation, precision, recall) Medium
A. One-hot encoding
B. Log transformation
C. Min-max scaling
D. Standardization

35 When is the F1-score preferred over accuracy as an evaluation metric?

Feature engineering and model evaluation (cross-validation, precision, recall) Medium
A. When only true negatives matter
B. When classes are imbalanced and both false positives and false negatives matter
C. When the dataset is perfectly balanced
D. When features are continuous

36 A disease affects 1% of a population. A test is 90% accurate for those with the disease and gives 10% false positives. If a person tests positive, what is the approximate probability they have the disease?

Bayes theorem Medium
A.
B.
C.
D.

37 In Bayes' theorem , what does represent?

Bayes theorem Medium
A. The evidence
B. The prior probability of
C. The posterior probability of
D. The likelihood of

38 In a Bayesian network, what does a directed edge from node to node typically represent?

Bayesian networks Medium
A. A correlation with no direction
B. That and are independent
C. A direct conditional dependency where 's probability depends on
D. That equals

39 What key property allows a Bayesian network to factorize a joint probability distribution compactly?

Bayesian networks Medium
A. Every node connects to every other node
B. Each variable is conditionally independent of its non-descendants given its parents
C. All variables are mutually independent
D. All variables share the same distribution

40 A Naive Bayes classifier makes which key assumption about features?

Bayes theorem, Bayesian networks, and probabilistic reasoning Medium
A. Features are conditionally independent given the class label
B. Features must be continuous
C. The prior is always uniform
D. Features are perfectly correlated

41 A disease affects of a population. A test has sensitivity (true positive rate) and specificity (true negative rate). If a person tests positive, what is the approximate probability they actually have the disease?

Bayes theorem Hard
A.
B.
C.
D.

42 A fraud classifier detects of true fraud cases but also flags legitimate transactions as fraud. What are its precision and recall respectively?

Feature engineering and model evaluation (cross-validation, precision, recall) Hard
A. Precision , Recall
B. Precision , Recall
C. Precision , Recall
D. Precision , Recall

43 In Principal Component Analysis, the principal components are derived from the eigenvectors of the covariance matrix. What does the eigenvalue associated with each eigenvector represent?

Linear algebra (applied focus) Hard
A. The number of samples in the dataset
B. The variance captured along that component direction
C. The correlation between two original features
D. The mean of the projected data

44 In reinforcement learning, an agent with discount factor receives rewards of at , at , and at . What is the discounted return from ?

Supervised, unsupervised, and reinforcement learning: concepts and real-world use Hard
A.
B.
C.
D.

45 A model's predictions on a test set have a systematic offset from the true values across all samples. In the bias-variance framework, this consistent error most directly reflects which quantity?

Statistics Hard
A. High irreducible noise
B. High bias
C. High variance
D. Low bias, high variance

46 Two events and satisfy , , and . Are and independent, and why?

Probability Hard
A. Yes, because
B. No, because
C. Yes, because
D. No, because

47 In a Bayesian network (a collider at ), which statement about conditional independence is correct?

Bayesian networks Hard
A. and are independent whether or not is observed
B. and become independent only when conditioning on
C. and are independent, but become dependent when conditioning on
D. and are always dependent regardless of

48 When performing -fold cross-validation, feature scaling parameters (e.g., mean and std) are computed on the entire dataset before splitting. Why is this problematic?

Feature engineering and model evaluation (cross-validation, precision, recall) Hard
A. It always increases the model's bias
B. It causes data leakage, inflating validation performance estimates
C. It makes the folds unequal in size
D. It reduces the number of usable folds

49 For a highly right-skewed income distribution, which measure of central tendency best represents the 'typical' value, and which will be largest?

Statistics Hard
A. Mean is best; mode is largest
B. Mode is best; median is largest
C. Median is best; mean is largest
D. Mean is best; median is largest

50 A square matrix used as a linear transformation has a determinant of . What does this imply about the transformation and solving ?

Linear algebra (applied focus) Hard
A. The transformation is non-invertible; has no unique solution
B. The transformation preserves volume; has one solution
C. The matrix is orthogonal; always solvable
D. The eigenvalues are all positive; is unique

51 A retailer wants to group customers by purchasing behavior without predefined labels, then use these groups to target marketing. Which learning paradigm fits the grouping task?

Supervised, unsupervised, and reinforcement learning: concepts and real-world use Hard
A. Supervised learning (classification)
B. Unsupervised learning (clustering)
C. Supervised learning (regression)
D. Reinforcement learning (policy optimization)

52 A biased coin lands heads with probability . You observe heads in tosses. Under a maximum likelihood estimate, what is , and what is a key limitation of this estimate?

Probability Hard
A. ; it correctly incorporates a Beta prior
B. ; it overfits small samples, ignoring uncertainty
C. ; it applies Laplace smoothing
D. ; it assumes a fair coin prior

53 A classifier achieves with precision . What is its recall?

Feature engineering and model evaluation (cross-validation, precision, recall) Hard
A.
B.
C.
D.

54 In a Naive Bayes classifier, the 'naive' assumption is that features are conditionally independent given the class. What is the primary consequence when features are actually strongly correlated?

Bayes theorem Hard
A. The classifier always predicts the majority class
B. Training time grows exponentially with correlation
C. Predicted class probabilities become poorly calibrated, though ranking may still work
D. The model cannot make any predictions at all

55 A hypothesis test yields at significance level . Which interpretation is correct?

Statistics Hard
A. Accept the null; the effect has a probability of being real
B. There is a chance the alternative hypothesis is false
C. The null hypothesis is true with probability
D. Reject the null; there is a chance of data this extreme if the null were true

56 In reinforcement learning, an agent that always chooses the action with the highest current estimated value faces which fundamental problem?

Supervised, unsupervised, and reinforcement learning: concepts and real-world use Hard
A. Insufficient exploration, potentially missing better long-term actions
B. Reward hacking due to a misspecified reward
C. Overfitting to the state space size
D. Excessive exploration, never converging on a policy

57 A Bayesian network has binary variables in a chain . How many independent parameters are needed to fully specify the joint distribution?

Bayesian networks Hard
A.
B.
C.
D.

58 In gradient descent, the update rule is . If the learning rate is set far too large, what typically happens?

Linear algebra (applied focus) Hard
A. The gradient becomes zero immediately
B. The model converges faster to the exact minimum
C. The loss diverges or oscillates instead of converging
D. The parameters remain unchanged across iterations

59 For a severely imbalanced dataset ( negative class), a model predicting 'negative' for everything achieves accuracy. Which metric best exposes this model's true weakness?

Feature engineering and model evaluation (cross-validation, precision, recall) Hard
A. Specificity
B. Recall on the positive class
C. Negative predictive value
D. Overall accuracy

60 A random variable has and . What is the standard deviation of ?

Probability Hard
A.
B.
C.
D.
Previous MCQ
Unit 2
Next MCQ
Unit 4