Data Mining MCQs

The two itemsets share the ordered prefix A and differ in the final item.
Their union forms the candidate {A, B, C}.

Although transactions are compressed, shared-prefix counts and node links retain relevant frequency information.
The original transaction list need not be reproduced.

Accuracy equals correct predictions divided by total predictions.
It can be misleading when classes are highly imbalanced.

Gini impurity is zero for a pure node.
It increases as class proportions become more mixed.

Naive Bayes factorizes the class-conditional likelihood across features.
The independence assumption is often unrealistic but computationally useful.

Among cases assigned probability 0.8, roughly 80 percent should be positive if calibrated.
A classifier may rank well while producing poorly calibrated probabilities.

This is the candidate prune step based on anti-monotonicity.
It prevents counting candidates that cannot possibly be frequent.

All filtered transaction paths begin under the root.
The root itself does not correspond to an item.

Precision measures the reliability of positive predictions.
It is important when false positives are costly.

Standard tree induction makes locally optimal choices at each node.
It does not usually search every possible complete tree.

Bayes' theorem combines prior class probability with feature likelihood.
The result is a posterior probability for each class.

Risk decisions depend on probability meaning, not just class ranking.
Calibration plots and proper scoring rules complement AUC.