What does a machine-learning model learn from examples?
A A fixed list of rules
B Only database table names
C Patterns in the data
D Random output sequences
Answer: C. Patterns in the data. A model uses examples to learn patterns that connect inputs with outputs or reveal structure in the data.
Which task predicts a numerical value, such as a house price?
A Regression
B Clustering
C Classification
D Ranking
Answer: A. Regression. Regression predicts a numerical value, such as a price or temperature.
Which learning type uses labeled examples?
A Unsupervised learning
B Reinforcement learning
C Transfer learning
D Supervised learning
Answer: D. Supervised learning. Supervised learning trains on examples that include the target label.
In a labeled dataset, what is the value a model must predict called?
A Feature
B Label
C Epoch
D Cluster
Answer: B. Label. The label is the target value the model learns to predict.
Which task sorts messages into spam or not spam?
A Regression
B Clustering
C Classification
D Compression
Answer: C. Classification. Spam detection assigns each message to a category, so it is classification.
What is a feature in a dataset?
A A model’s final score
B A training-time reward
C A predicted category
D An input used for prediction
Answer: D. An input used for prediction. A feature is an input variable that may help predict the target.
Which dataset portion is used to fit model parameters?
A Training set
B Test set
C Holdout report
D Prediction log
Answer: A. Training set. The training set supplies examples used to fit the model’s parameters.
What does a test set mainly measure?
A Training speed
B Feature count
C Performance on unseen examples
D Dataset size
Answer: C. Performance on unseen examples. A separate test set estimates how well the trained model performs on unseen examples.
Which method groups similar unlabeled records?
A Regression
B Clustering
C Classification
D Ranking
Answer: B. Clustering. Clustering groups examples by similarity without requiring target labels.
When should a reinforcement-learning agent receive a reward?
A Before observing anything
B Only during data cleaning
C Only after a test split
D After an action or outcome
Answer: D. After an action or outcome. A reward gives feedback about an agent’s action or resulting state.
Which output suits a regression task?
A A continuous number
B A class name
C A group identifier
D A yes-or-no label
Answer: A. A continuous number. Regression is used when the target is numerical and can vary over a range.
What does one epoch mean during model training?
A One prediction for one row
B One pass through training data
C One category split
D One test-set relabeling
Answer: B. One pass through training data. An epoch is one full pass through the training examples.
Which item is commonly a hyperparameter?
A The target label
B A test prediction
C The learning rate
D A missing value
Answer: C. The learning rate. The learning rate is set before or during model tuning rather than learned as a model parameter.
Which split is used to tune model choices?
A Validation set
B Training label
C Deployment log
D Final report
Answer: A. Validation set. The validation set helps compare model choices during development.
What does overfitting usually mean?
A The model has no features
B The model predicts only averages
C Too little training data
D Memorizes training noise
Answer: D. Memorizes training noise. An overfit model captures details of training data that do not carry over to new examples.
Which algorithm predicts from nearby training examples?
A Naive Bayes
B K-nearest neighbors
C Linear regression
D K-means
Answer: B. K-nearest neighbors. K-nearest neighbors uses the labels or values of nearby examples to make a prediction.
Which measure is the fraction of correct classifications?
A Recall
B Precision
C Accuracy
D Specificity
Answer: C. Accuracy. Accuracy is the number of correct predictions divided by the total number of predictions.
What is missing from an unlabeled dataset?
A Input features
B Example rows
C A training split
D Target labels
Answer: D. Target labels. Unlabeled data has input examples but no provided target labels.
Which technique penalizes model complexity to limit overfitting?
A Shuffling
B Regularization
C Encoding
D Sampling
Answer: B. Regularization. Regularization adds a complexity penalty to discourage overly complex models.
Given actual value 8 and prediction 6, what is the absolute error?
A 2
B 0
C 1
D 3
Answer: A. 2. Absolute error is the absolute difference: |8 − 6| = 2.
What does a loss function measure?
A The model’s storage size
B The dataset’s age
C The number of features
D Prediction error
Answer: D. Prediction error. A loss function quantifies the difference between predictions and target values.
Which model can split data using feature-based questions?
A K-means
B PCA
C Decision tree
D Linear regression
Answer: C. Decision tree. A decision tree applies feature-based tests at nodes to split examples.
Why might test accuracy be low when training accuracy is high?
A The model may be overfitting
B The labels are always correct
C The test set is larger
D The model has no parameters
Answer: A. The model may be overfitting. A large gap between training and test performance often signals overfitting.
Which example is a binary-classification task?
A Estimate tomorrow’s temperature
B Predict loan default
C Predict monthly sales
D Group photos by similarity
Answer: B. Predict loan default. The output has two classes: default or not default.
Which setting helps repeat a randomized experiment?
A A larger font size
B A new target label
C A fixed random seed
D A lower file name
Answer: C. A fixed random seed. Using the same random seed can make many randomized steps repeatable in the same setup.