Question 601
A cluster that contains points with very different densities and shapes may be better handled by:
A. DBSCAN
B. K‑means
C. Linear regression
D. Logistic regression
Check answer
Select an option first.
Correct answer: A — DBSCAN
Explanation: Why A is right: DBSCAN handles arbitrary shapes and densities. Why others are wrong: B: Assumes spherical clusters. C/D: Not clustering algorithms.
Question 602
The main advantage of DBSCAN over K‑means is that DBSCAN:
A. Can detect clusters of arbitrary shape
B. Requires specifying k
C. Uses centroids
D. Requires normalized data
Check answer
Select an option first.
Correct answer: A — Can detect clusters of arbitrary shape
Explanation: Why A is right: DBSCAN identifies clusters of any shape and handles noise well. Why others are wrong: B: DBSCAN does not require k. C: No centroids. D: Scaling helps but is not a requirement.
Question 603
Adjusted (R^2) is used instead of (R^2) because:
A. It ignores the number of variables
B. It penalizes adding variables that do not improve the model
C. It is always larger than (R^2)
D. It is easier to compute
Check answer
Select an option first.
Correct answer: B — It penalizes adding variables that do not improve the model
Explanation: Right: Adjusted (R^2) accounts for the number of predictors. Wrong: A: Opposite. C: It can be smaller. D: It is slightly more complex.
Question 604
A model that fits the historical data extremely well but performs poorly on new data is likely:
A. Underfitted
B. Overfitted
C. Perfectly specified
D. Free of errors
Check answer
Select an option first.
Correct answer: B — Overfitted
Explanation: Right: Overfitting captures noise rather than true patterns. Wrong: A/C/D: Not consistent with poor out-of-sample performance.
Question 605
A predicted probability from a logistic regression of 0.8 for default means:
A. Default is guaranteed
B. Default is impossible
C. There is an 80% estimated chance of default
D. The firm will default exactly 80% of the time
Check answer
Select an option first.
Correct answer: C — There is an 80% estimated chance of default
Explanation: Right: 0.8 is an estimated probability, not a certainty. Wrong: A/B/D: Misinterpret probabilities.
Get the full CFA question bank — free
Drop your email and we'll send you fresh CFA practice questions, fully worked solutions and exam-deadline reminders. No spam — unsubscribe in one click.
Question 606
Supervised learning involves:
A. No labeled data
B. Labeled input–output pairs
C. Only text data
D. Only image data
Check answer
Select an option first.
Correct answer: B — Labeled input–output pairs
Explanation: Right: Supervised learning uses labeled examples. Wrong: A/C/D: Not defining features.
Question 607
Overfitting in machine learning means:
A. The model is too simple
B. The model captures noise in the training data and performs poorly on new data
C. The model has no parameters
D. The model has no training data
Check answer
Select an option first.
Correct answer: B — The model captures noise in the training data and performs poorly on new data
Explanation: Right: Overfitting = memorizing noise. Wrong: A/C/D: Not overfitting.
Question 608
A common way to check for overfitting is to:
A. Evaluate performance only on training data
B. Evaluate performance on separate test or validation data
C. Ignore performance metrics
D. Use only one observation
Check answer
Select an option first.
Correct answer: B — Evaluate performance on separate test or validation data
Explanation: Right: Test/validation sets reveal overfitting. Wrong: A/C/D: Not sufficient.
Question 609
Penalized regression methods (like Lasso or Ridge) are used to:
A. Increase model complexity
B. Reduce overfitting by shrinking coefficients
C. Remove the intercept
D. Eliminate residuals
Check answer
Select an option first.
Correct answer: B — Reduce overfitting by shrinking coefficients
Explanation: Right: Penalties shrink coefficients and control complexity. Wrong: A/C/D: Not their purpose.
Question 610
A decision tree model:
A. Splits data into regions based on feature values
B. Always produces linear decision boundaries
C. Cannot handle categorical variables
D. Has no structure
Check answer
Select an option first.
Correct answer: A — Splits data into regions based on feature values
Explanation: Right: Trees recursively split the feature space. Wrong: B/C/D: Incorrect.
Question 611
A random forest is:
A. A single decision tree
B. An ensemble of decision trees built on different samples or features
C. A linear regression model
D. A clustering algorithm
Check answer
Select an option first.
Correct answer: B — An ensemble of decision trees built on different samples or features
Explanation: Right: Random forests aggregate many trees. Wrong: A/C/D: Different methods.
Question 612
In supervised learning, a “training set” is:
A. Data used to evaluate final performance
B. Data used to fit the model
C. Data that is never used
D. Data with no features
Check answer
Select an option first.
Correct answer: B — Data used to fit the model
Explanation: Right: Training data is used to estimate model parameters. Wrong: A/C/D: Not correct.
Question 613
A “test set” is used to:
A. Fit the model
B. Tune hyperparameters
C. Evaluate model performance on unseen data
D. Remove outliers
Check answer
Select an option first.
Correct answer: C — Evaluate model performance on unseen data
Explanation: Right: Test sets assess generalization. Wrong: A/B/D: Different roles.
Question 614
Cross-validation is a technique to:
A. Randomly shuffle labels
B. Assess model performance by repeatedly splitting data into training and validation sets
C. Remove all residuals
D. Force the model to be linear
Check answer
Select an option first.
Correct answer: B — Assess model performance by repeatedly splitting data into training and validation sets
Explanation: Right: Cross-validation evaluates stability and performance. Wrong: A/C/D: Not its purpose.
Question 615
A neural network consists of:
A. Only one linear equation
B. Layers of interconnected nodes (neurons)
C. Only a single decision tree
D. Only a moving average
Check answer
Select an option first.
Correct answer: B — Layers of interconnected nodes (neurons)
Explanation: Right: Neural nets are layered networks of nodes. Wrong: A/C/D: Different models.
Question 616
Deep learning refers to neural networks with:
A. No hidden layers
B. Many hidden layers
C. Only one neuron
D. Only one output
Check answer
Select an option first.
Correct answer: B — Many hidden layers
Explanation: Right: “Deep” = multiple hidden layers. Wrong: A/C/D: Not deep.
Question 617
Feature selection aims to:
A. Use all available variables regardless of relevance
B. Choose a subset of variables that are most useful for the model
C. Remove the dependent variable
D. Randomly drop observations
Check answer
Select an option first.
Correct answer: B — Choose a subset of variables that are most useful for the model
Explanation: Right: Feature selection focuses on relevant predictors. Wrong: A/C/D: Not the goal.
Question 618
If two independent variables are uncorrelated, this means:
A. They cannot both be useful predictors
B. They may each explain different parts of the variation in the dependent variable
C. The model is misspecified
D. The intercept must be zero
Check answer
Select an option first.
Correct answer: B — They may each explain different parts of the variation in the dependent variable
Explanation: Uncorrelated predictors can each add value. A/C/D: Not implied.
Question 619
A model with too many variables may show:
A. Overfitting
B. Underfitting
C. Zero residuals
D. Perfect predictions
Check answer
Select an option first.
Correct answer: A — Overfitting
Explanation: Too many predictors → overfitting. B/C/D: Incorrect.
Question 620
A model with too few variables may show:
A. Underfitting
B. Overfitting
C. Zero residuals
D. Perfect predictions
Check answer
Select an option first.
Correct answer: A — Underfitting
Explanation: Missing predictors → underfitting. B/C/D: Incorrect.
Question 621
A model that fits the training data extremely well but fails on new data is:
A. Underfitted
B. Overfitted
C. Perfect
D. Free of errors
Check answer
Select an option first.
Correct answer: B — Overfitted
Explanation: Overfitting captures noise. A/C/D: Incorrect.
Question 622
A key difference between supervised and unsupervised learning is:
A. Supervised uses labeled data; unsupervised does not
B. Both use labeled data
C. Both use no data
D. Both require binary outcomes
Check answer
Select an option first.
Correct answer: A — Supervised uses labeled data; unsupervised does not
Explanation: Labels distinguish supervised learning. B/C/D: Incorrect.
Question 623
A model that memorizes training data but fails on new data is:
A. Underfitted
B. Overfitted
C. Perfect
D. Free of errors
Check answer
Select an option first.
Correct answer: B — Overfitted
Explanation: Overfitting = memorizing noise. A/C/D: Incorrect.
Question 624
A validation set is used to:
A. Fit the model
B. Tune hyperparameters
C. Evaluate final performance
D. Remove outliers
Check answer
Select an option first.
Correct answer: B — Tune hyperparameters
Explanation: Validation data tunes model settings. A/C/D: Incorrect.
Question 625
A test set is used to:
A. Fit the model
B. Tune hyperparameters
C. Evaluate final model performance
D. Remove missing values
Check answer
Select an option first.
Correct answer: C — Evaluate final model performance
Explanation: Test data evaluates generalization. A/B/D: Incorrect.
Question 626
Regularization methods help reduce:
A. Underfitting
B. Overfitting
C. Residuals
D. Intercepts
Check answer
Select an option first.
Correct answer: B — Overfitting
Explanation: Regularization controls complexity. A/C/D: Incorrect.
Question 627
A decision tree may overfit because:
A. It is too shallow
B. It grows too deep
C. It has no parameters
D. It uses no data
Check answer
Select an option first.
Correct answer: B — It grows too deep
Explanation: Deep trees memorize noise. A/C/D: Incorrect.
Question 628
A random forest reduces overfitting by:
A. Using one tree
B. Averaging many trees
C. Removing all features
D. Using no randomness
Check answer
Select an option first.
Correct answer: B — Averaging many trees
Explanation: Averaging reduces variance. A/C/D: Incorrect.
Question 629
Feature engineering involves:
A. Creating new variables that improve model performance
B. Removing the dependent variable
C. Randomly changing values
D. Sorting data
Check answer
Select an option first.
Correct answer: A — Creating new variables that improve model performance
Explanation: Feature engineering enhances predictive power. B/C/D: Incorrect. A. Trend Models (Linear vs Log‑Linear)
Question 630
ARCH behavior implies that variance is:
A. Constant
B. Predictable based on past shocks
C. Zero
D. Unrelated to past values
Check answer
Select an option first.
Correct answer: B — Predictable based on past shocks
Explanation: Why right: ARCH models variance as a function of past squared errors. Wrong: A/C/D: Incorrect.
More free CFA topics