Data Mining MCQs

Large neighboring distances can indicate cluster boundaries.
Small distances suggest homogeneous regions of the map.

Recall measures the fraction of actual anomalies detected.
High recall is vital when false negatives are costly.

SVM seeks a boundary that leaves the widest gap between classes.
A larger margin is associated with better robustness.

Nearby or strongly connected observations are assumed likely to share labels.
The graph structure helps exploit unlabeled data.

An empty cluster has no points from which to compute a mean.
Algorithms handle this by relocating or replacing the centroid.

SOM aims to retain neighborhood relationships rather than exact geometry.
This makes the map useful for visual exploration.

Predicting every case as normal can achieve high accuracy.
Minority-focused metrics provide a more meaningful evaluation.

Support vectors lie on or inside the margin boundaries.
Moving non-support-vector points slightly often does not change the classifier.

Unlabeled observations reveal the geometry and density of the input space.
Labels can then be extended along that structure.

Agglomerative clustering repeatedly merges the closest clusters.
The process forms a bottom-up hierarchy.

Quantization error measures how well prototypes represent the observations.
Lower values indicate closer representation, though map complexity also matters.

Different thresholds trade false positives against missed anomalies.
The right choice depends on risk, capacity, and class prevalence.