Data Mining MCQs

Support is computed directly from observed frequencies.
Usefulness and novelty depend partly on stakeholder goals and prior knowledge.

Normalization prevents large-scale variables from dominating scale-sensitive algorithms.
Common methods include min-max scaling and z-score standardization.

Feature selection reduces dimensionality without transforming the retained variables.
It can improve speed, interpretability, and generalization.

The consequent occurs less often with the antecedent than expected from its baseline.
Such a rule describes avoidance or negative dependence.

Constraints can restrict items, rule length, aggregate values, or business conditions.
They reduce search and focus the output on relevant patterns.

Any globally frequent itemset must be frequent in at least one partition.
This property limits global candidates and can reduce database scans.

Domain knowledge guides feature selection, constraints, and practical interpretation.
Without context, an algorithm may find patterns that are statistically real but irrelevant.

Min-max scaling uses the observed minimum and maximum.
It preserves order but can be sensitive to extreme values.

Leakage creates unrealistically high evaluation results.
Examples include preprocessing on the full dataset or using future information.

Leverage is support(XY) minus support(X)support(Y).
Zero leverage corresponds to independence.

The Apriori property follows from support anti-monotonicity.
An itemset cannot occur more often than any of its subsets.

Sampling reduces computation by working on fewer transactions.
A verification step may be needed because some true patterns can be missed.