Unit 13: Machine learning - Practice Quiz

ECAP776 60 Questions
0 Correct 0 Wrong 60 Left
0/60

1 What is machine learning?

Introduction Easy
A. A method that enables computers to learn patterns from data
B. A system that stores files without processing them
C. A method used only to repair computer hardware
D. A language designed for creating web page layouts

2 What is the main purpose of training a machine learning model?

Introduction Easy
A. To delete incorrect files from a computer
B. To increase the physical size of memory
C. To learn useful patterns from available data
D. To convert every program into machine code

3 What is a machine learning model?

Introduction Easy
A. A spreadsheet containing only empty cells
B. A physical device used to collect data
C. A learned representation used to make predictions
D. A network cable connecting two computers

4 Which data is used to teach a machine learning model?

Introduction Easy
A. Archive data
B. Training data
C. Backup data
D. Display data

5 In machine learning, what is a feature?

Introduction Easy
A. An input characteristic used by a model
B. A final score assigned after testing
C. A password required to open a dataset
D. A hardware component inside a processor

6 What is a label in a labeled dataset?

Introduction Easy
A. The location where the dataset is stored
B. The program used to open the dataset
C. The number of columns in the dataset
D. The known answer linked to an example

7 Why is test data used in machine learning?

Introduction Easy
A. To translate Python into another language
B. To increase the computer's screen size
C. To replace the model with a database
D. To evaluate the model on unseen examples

8 Which Python library is commonly used for basic machine learning tasks?

Introduction Easy
A. Pygame
B. scikit-learn
C. Tkinter
D. Turtle

9 What does a prediction represent in machine learning?

Introduction Easy
A. The output produced by a trained model
B. The folder containing the Python program
C. The physical connection between two devices
D. The process of installing a software package

10 Which example is a common application of machine learning?

Introduction Easy
A. Connecting a monitor with a video cable
B. Changing a keyboard's physical layout
C. Printing the same fixed sentence repeatedly
D. Detecting spam in incoming email messages

11 Which type of machine learning uses labeled training data?

Types of machine learning Easy
A. Supervised learning
B. Reinforcement learning
C. Unsupervised learning
D. Random learning

12 Which type of machine learning discovers patterns in unlabeled data?

Types of machine learning Easy
A. Manual learning
B. Unsupervised learning
C. Reinforcement learning
D. Supervised learning

13 In which type of machine learning does an agent learn through rewards and penalties?

Types of machine learning Easy
A. Sequential searching
B. Reinforcement learning
C. Supervised learning
D. Unsupervised learning

14 Which task predicts a category such as "spam" or "not spam"?

Types of machine learning Easy
A. Clustering
B. Regression
C. Classification
D. Compression

15 Which task predicts a continuous numerical value such as a house price?

Types of machine learning Easy
A. Association
B. Classification
C. Regression
D. Clustering

16 Which task groups similar data items without using known labels?

Types of machine learning Easy
A. Classification
B. Regression
C. Clustering
D. Prediction

17 Predicting whether a student will pass or fail is an example of which task?

Types of machine learning Easy
A. Sorting
B. Classification
C. Regression
D. Clustering

18 Grouping customers by similar purchasing behavior is an example of which task?

Types of machine learning Easy
A. Regression
B. Classification
C. Calculation
D. Clustering

19 A robot learning to move by receiving points for successful actions uses which type of learning?

Types of machine learning Easy
A. Supervised learning
B. Database learning
C. Reinforcement learning
D. Unsupervised learning

20 Predicting tomorrow's temperature as a number is an example of which task?

Types of machine learning Easy
A. Clustering
B. Regression
C. Classification
D. Association

21 A house-price model performs very well on its training data but poorly on newly collected houses. Which problem does this most strongly indicate?

Introduction Medium
A. Overfitting
B. Feature extraction
C. Data normalization
D. Underfitting

22 A dataset contains customer age, annual income, and whether each customer repaid a loan. When predicting repayment, what is the role of annual income?

Introduction Medium
A. A label
B. A loss value
C. A prediction
D. A feature

23 Why should a test dataset normally remain separate from the data used to train a model?

Introduction Medium
A. To estimate performance on unseen examples
B. To increase the number of model parameters
C. To ensure every feature has the same value
D. To repeatedly adjust the model until it memorizes all available examples

24 A company trains a model using emails already labeled as "spam" or "not spam." Which type of machine learning is being used?

Types of machine learning Medium
A. Supervised learning
B. Self-organizing learning
C. Reinforcement learning
D. Unsupervised learning

25 A retailer groups customers according to purchasing behavior without providing predefined customer categories. Which approach is most appropriate?

Types of machine learning Medium
A. Clustering
B. Classification
C. Reinforcement learning
D. Regression

26 An agent learns to navigate a maze by receiving a positive score at the exit and penalties for collisions. Which learning type does this illustrate?

Types of machine learning Medium
A. Unsupervised learning
B. Supervised learning
C. Dimensionality reduction
D. Reinforcement learning

27 A medical dataset records patient temperature in Celsius and body weight in kilograms. Why might feature scaling be useful before training some models?

Introduction Medium
A. It gives features comparable numerical ranges
B. It removes every incorrect record
C. It converts all inputs into target labels
D. It guarantees that the resulting model will achieve perfect accuracy on future patients

28 Which task is an example of supervised regression?

Types of machine learning Medium
A. Predicting tomorrow's temperature
B. Choosing actions from rewards
C. Grouping articles by similarity
D. Assigning photos to unlabeled clusters

29 Which task is best formulated as supervised classification?

Types of machine learning Medium
A. Compressing many features into two numerical dimensions while preserving as much variation as possible
B. Grouping songs without genre labels
C. Predicting a car's selling price
D. Determining whether a transaction is fraudulent

30 A model achieves an accuracy of on 500 test examples. Approximately how many examples did it classify correctly?

Introduction Medium
A. 336
B. 484
C. 400
D. 420

31 Which change is most likely to improve a model that underfits both its training data and validation data?

Introduction Medium
A. Evaluate only on training data
B. Replace the validation labels with the model's own predictions before measuring performance
C. Use a more expressive model
D. Remove most training examples

32 A column containing unique student identification numbers is included when predicting exam performance. Why might this column be removed?

Introduction Medium
A. It is necessarily the target being predicted
B. It is usually not a meaningful predictive feature
C. It always contains missing numerical values
D. It automatically transforms the problem from supervised learning into reinforcement learning

33 A school has 10,000 student records, but only 800 have been labeled as "at risk" or "not at risk." Which learning approach is designed to use both sets?

Types of machine learning Medium
A. Hierarchical clustering
B. Semi-supervised learning
C. Pure supervised learning
D. Reinforcement learning

34 A model learns useful image representations by predicting hidden portions of unlabeled images before being adapted to classify objects. What approach is this?

Types of machine learning Medium
A. Fully supervised regression using manually supplied continuous labels for every image
B. Standard clustering
C. Self-supervised learning
D. Reinforcement learning

35 During model development, what is the primary purpose of a validation set?

Introduction Medium
A. To store deployed predictions
B. To replace all training examples
C. To select model settings
D. To provide the final unbiased performance report after being checked repeatedly during development

36 Two input features contain the same information: age in years and age in months. What concern does this create?

Introduction Medium
A. The test set automatically becomes larger
B. The labels become unsupervised
C. The task changes into clustering
D. The features are strongly redundant

37 A streaming service must recommend an item, observe whether the user engages with it, and adapt future recommendations. Which framework best captures this repeated interaction?

Types of machine learning Medium
A. Reinforcement learning
B. Supervised regression
C. Dimensionality reduction followed by a fixed rule that never responds to user feedback
D. Static clustering

38 Which statement correctly distinguishes classification from clustering?

Types of machine learning Medium
A. Classification requires rewards; clustering requires penalties
B. Classification predicts numbers; clustering predicts known labels
C. Classification always uses fewer records than clustering because every category must contain one example
D. Classification uses known categories; clustering discovers groups

39 A fraud dataset contains 9,900 legitimate transactions and 100 fraudulent transactions. A model predicts every transaction as legitimate. Why is its accuracy misleading?

Introduction Medium
A. It uses too many numerical features
B. It proves that accuracy can only be used when both classes contain exactly the same number of records
C. It was necessarily trained without labels
D. It fails to identify any fraudulent transactions

40 A trained model receives a new customer's attributes and outputs a churn probability. What is this use of the trained model called?

Introduction Medium
A. Clustering
B. Inference
C. Hyperparameter tuning
D. Labeling

41 A team evaluates 100 model configurations on the same validation set and reports the configuration with the highest validation accuracy as its final performance. Even though no model was trained on that validation set, why is the reported accuracy likely optimistic?

Introduction Hard
A. Repeated selection has indirectly overfit the validation set
B. Validation accuracy is always higher than test accuracy
C. The configurations necessarily share identical parameters
D. Training without validation examples causes underfitting

42 A dataset contains 500 labeled radiographs and 50,000 unlabeled radiographs. A model first learns representations by predicting masked regions in all images and is then fine-tuned using the 500 labels. Which description is most precise?

Types of machine learning Hard
A. Self-supervised pretraining followed by supervised fine-tuning
B. Reinforcement learning followed by transductive inference
C. Semi-supervised clustering followed by active learning with repeated expert queries
D. Supervised pretraining followed by unsupervised clustering

43 A binary classifier is evaluated on a population where the positive class prevalence changes from during testing to after deployment, while both and the decision threshold remain unchanged. Which metric is guaranteed to remain unchanged in expectation?

Introduction Hard
A. Accuracy
B. Precision
C. Negative predictive value
D. Recall

44 An agent observes a context, chooses one advertisement, and immediately receives a click or no-click reward. Its action does not affect the next context. What is the most specific learning formulation?

Types of machine learning Hard
A. Supervised learning with fixed labels collected independently of the policy
B. Contextual bandit learning
C. Full sequential reinforcement learning
D. Unsupervised density estimation

45 A regression problem follows , where and . With squared-error loss and unlimited representative data, what prevents the expected prediction error for from reaching zero?

Introduction Hard
A. The irreducible conditional noise
B. The absence of a classification threshold
C. The use of continuous target values
D. The model's inability to reproduce every training example exactly

46 A learner receives the complete unlabeled test feature set together with labeled training data. It may tailor predictions jointly to those particular test points but is not required to produce a rule for future points. Which setting is this?

Types of machine learning Hard
A. Transductive learning
B. Inductive learning
C. Online reinforcement learning
D. Self-supervised representation learning over an indefinitely expanding population

47 A fraud detector has excellent cross-validation results, but transaction records from the same customer appear in both training and validation folds. The deployment goal is prediction for entirely new customers. Which correction best aligns validation with deployment?

Introduction Hard
A. Remove all customer-level features
B. Balance fraud prevalence independently inside every randomly formed record-level fold
C. Split records by customer identity
D. Increase the number of randomly generated folds

48 A clustering algorithm groups customer records without labels. Analysts later assign business names to the clusters after inspecting them. How should the algorithm's original training be classified?

Types of machine learning Hard
A. Semi-supervised, because humans interpreted its output
B. Reinforcement-based, because analysts accepted or rejected the discovered customer groups
C. Supervised, because the final clusters receive names
D. Unsupervised, because no target labels guided fitting

49 Suppose training and deployment satisfy but . What kind of distribution change is this?

Introduction Hard
A. Label shift
B. A change in the annotation policy that alters both class definitions and conditional outcomes
C. Concept drift
D. Covariate shift

50 A spam filter asks a human to label only the emails for which its current prediction is most uncertain, then retrains on the expanded labeled set. Which paradigm primarily describes this label-acquisition strategy?

Types of machine learning Hard
A. Active learning
B. Transfer learning
C. Reinforcement learning with a long-horizon reward assigned after every labeling campaign
D. Unsupervised learning

51 A model attains zero training error after unique identification numbers are included as numeric features. Why does this result alone provide almost no evidence of useful learning?

Introduction Hard
A. Numeric features cannot represent categorical identifiers
B. Zero training error proves that regularization was excessive
C. Any feature unique to each observation forces every possible learning algorithm to predict randomly on deployment data
D. Identification numbers may enable memorization without transferable structure

52 A model predicts whether each image contains any of the labels {car, pedestrian, bicycle}, allowing several labels to be true simultaneously. Which task type is most precise?

Types of machine learning Hard
A. Structured reinforcement learning in which each object label represents an available action
B. Multi-label classification
C. Multiclass classification
D. Ordinal regression

53 A feature-standardization transform is fitted once using the entire dataset and then cross-validation is performed on the transformed data. What is the central methodological error?

Introduction Hard
A. The transform guarantees that each validation fold has exactly zero mean and unit variance independently
B. Standardization changes regression into classification
C. Cross-validation requires all features to remain unscaled
D. The transform exposes validation-fold statistics during training

54 Which statement correctly relates online learning and supervised learning?

Types of machine learning Hard
A. Every online method must update exactly once per example and permanently discard all previous parameters
B. Supervised learning is online only when it uses reinforcement rewards
C. They are mutually exclusive because online learning has no labels
D. Online learning can be supervised when labeled examples arrive sequentially

55 Two models have identical average test accuracy. Model A is correct on nearly all examples from one subgroup and poor on another, while Model B performs moderately on both. What follows logically from the identical average?

Introduction Hard
A. Model A necessarily has lower variance than Model B
B. Their deployment consequences may still differ substantially
C. Their subgroup error rates must also be identical
D. Model B is guaranteed to be optimal under every possible operational cost matrix

56 A neural network is trained to distinguish real samples from generated samples while another network learns to generate samples that fool it. No human class labels are used. Which description best fits the overall task?

Types of machine learning Hard
A. Supervised multiclass classification
B. Semi-supervised discriminative learning because the real and generated indicators constitute permanent human-provided target classes
C. Contextual bandit optimization
D. Unsupervised generative learning

57 Under the no-free-lunch principle, why can no learning algorithm be universally superior over all possible prediction problems?

Introduction Hard
A. All algorithms produce identical models on finite datasets
B. Generalization becomes impossible whenever labels contain noise
C. Any advantage relies on assumptions about the problem distribution
D. Every universally consistent learner must have the same finite-sample error on each individual dataset

58 A robot receives a reward only after completing a 30-action assembly sequence. Which issue is characteristic of reinforcement learning and is not captured by ordinary independently labeled supervised examples?

Types of machine learning Hard
A. Class balancing across target categories
B. Feature scaling across numeric inputs
C. Credit assignment across earlier actions
D. Selecting a deterministic train-test split that preserves every action sequence in both partitions

59 A classifier is trained from labels produced by several imperfect heuristics rather than directly verified human annotations. The heuristics sometimes conflict. Which paradigm most specifically describes the supervision source?

Types of machine learning Hard
A. Weak supervision
B. Pure unsupervised learning
C. On-policy reinforcement learning
D. Self-supervised learning in which every heuristic label is guaranteed to be a deterministic transformation of the raw input

60 Before cross-validation, an analyst applies PCA to all available feature vectors without using target labels. Why can this still invalidate the estimated generalization performance?

Introduction Hard
A. Cross-validation permits dimensionality reduction only when every original feature is statistically independent
B. PCA necessarily removes the target-predictive directions
C. Validation features influenced the learned projection
D. PCA is supervised whenever evaluation labels exist