Practice Library
All MCQs
Browse exam-wise, subject-wise, and country-wise MCQs with explanations.
Choose an option to check your answer.
A.
An alternative split that approximates the chosen split when its feature is missing
B.
A split generated from the target label
C.
A rule with zero information gain
D.
A branch used only for pruning
Show Answer
Correct Answer: A. An alternative split that approximates the chosen split when its feature is missing
Explanation:
Surrogate splits preserve routing when the primary split value is unavailable.
Choose an option to check your answer.
A.
By converting every value to a class label manually
B.
By using support counts
C.
By ignoring the feature
D.
By testing thresholds such as x ≤ t
Show Answer
Correct Answer: D. By testing thresholds such as x ≤ t
Explanation:
Candidate thresholds divide ordered numeric values into regions.
Choose an option to check your answer.
A.
Replace entropy with Euclidean distance
B.
Increase tree depth automatically
C.
Adjust information gain for the intrinsic information of a split
D.
Calculate rule confidence
Show Answer
Correct Answer: C. Adjust information gain for the intrinsic information of a split
Explanation:
Gain ratio penalizes splits that fragment data into many small groups.
Choose an option to check your answer.
A.
It favors only binary attributes
B.
It may favor attributes with many distinct values
C.
It ignores class distributions
D.
It always selects the least useful feature
Show Answer
Correct Answer: B. It may favor attributes with many distinct values
Explanation:
A high-cardinality attribute can create many pure but tiny partitions.
Choose an option to check your answer.
A.
Splits depend on order and thresholds rather than distance magnitude
B.
Trees accept only categorical variables
C.
Scaling destroys class labels
D.
All tree features must have variance one
Show Answer
Correct Answer: A. Splits depend on order and thresholds rather than distance magnitude
Explanation:
Multiplying a feature by a positive constant changes threshold values but not ordering.
Choose an option to check your answer.
A.
By requiring a linear equation only
B.
By using no features
C.
By sorting classes alphabetically
D.
By combining multiple axis-aligned splits
Show Answer
Correct Answer: D. By combining multiple axis-aligned splits
Explanation:
A sequence of feature thresholds partitions the space into rectangular regions.
Choose an option to check your answer.
A.
It cannot represent nonlinear relationships
B.
It requires standardized features
C.
It can create leaves for noise and rare training cases
D.
It always predicts the majority class
Show Answer
Correct Answer: C. It can create leaves for noise and rare training cases
Explanation:
Deep branches may capture accidental patterns unique to training data.
Choose an option to check your answer.
A.
Stopping at the root
B.
Growing a larger tree and then removing weak subtrees
C.
Normalizing features after training
D.
Changing class labels after prediction
Show Answer
Correct Answer: B. Growing a larger tree and then removing weak subtrees
Explanation:
Post-pruning evaluates whether replacing subtrees with simpler leaves improves estimated generalization.
Choose an option to check your answer.
A.
Stopping tree growth before leaves become overly specific
B.
Growing a full tree and trimming it later
C.
Removing training examples
D.
Pruning input features before collecting data
Show Answer
Correct Answer: A. Stopping tree growth before leaves become overly specific
Explanation:
Pre-pruning uses conditions such as maximum depth or minimum node size.
Choose an option to check your answer.
A.
Adding every possible feature split
B.
Converting the tree to a graph database
C.
Deleting the target class
D.
Removing branches that add complexity without sufficient generalization benefit
Show Answer
Correct Answer: D. Removing branches that add complexity without sufficient generalization benefit
Explanation:
Pruning reduces overfitting by simplifying weak or noisy branches.
Choose an option to check your answer.
A.
They have no parameters
B.
They always contain one node
C.
Predictions can be traced through explicit feature-based rules
D.
They use only binary data
Show Answer
Correct Answer: C. Predictions can be traced through explicit feature-based rules
Explanation:
A root-to-leaf path forms an understandable if-then rule.
Choose an option to check your answer.
A.
The split guaranteeing the globally smallest tree
B.
The split giving the best immediate impurity reduction
C.
A random split in every case
D.
The split using the last feature
Show Answer
Correct Answer: B. The split giving the best immediate impurity reduction
Explanation:
Standard tree induction makes locally optimal choices at each node.
