Density-Based CLustering Part 2

What is the main idea of density-based clustering?

Find dense groups of points separated by sparse regions.

What does DBSCAN stand for?

Density-Based Spatial Clustering of Applications with Noise.

What does epsilon (ε) control in DBSCAN?

The maximum radius of a point's neighborhood.

What does MinPts control?

The minimum number of points required in an epsilon-neighborhood for a point to be considered a core point.

Does a point count itself toward MinPts?

Yes, the point itself is included when counting neighbors.

What is a core point?

A point with at least MinPts points in its epsilon-neighborhood, including itself.

What is a border point?

A point that does not satisfy MinPts but is within epsilon distance of a core point.

What is a noise point or outlier?

A point that is neither a core point nor a border point.

Which points can expand a DBSCAN cluster?

Core points, because they have enough neighbors to form a dense region.

Can a border point belong to a cluster?

Yes, but it cannot expand the cluster because it does not satisfy MinPts.

What is direct density reachability?

A point is directly density-reachable from another point if the starting point is a core point and the other point is within its epsilon-neighborhood.

Is direct density reachability symmetric?

Not always. A border point can be directly density-reachable from a core point, but the reverse is not necessarily true.

What is density reachability?

Two points are connected through a chain of directly density-reachable points.

Can density-reachable points be farther apart than epsilon?

Yes, as long as a valid chain of directly density-reachable points connects them.

Is density reachability always symmetric?

No. A border point can be reached from a core point but cannot necessarily reach that core point.

What is density connectivity?

Two points are density-connected if both are density-reachable from a common point.

Is density connectivity symmetric?

Yes. If A is density-connected to B, then B is density-connected to A.

What usually happens when epsilon increases?

More points may become core points, clusters may merge, and fewer points may be classified as noise.

What usually happens when epsilon decreases?

Fewer points qualify as core points, clusters may split, and more points may become noise.

What happens when MinPts increases?

It becomes harder for points to qualify as core points, which may increase the number of noise points.

What happens when MinPts decreases?

It becomes easier for points to qualify as core points, which may allow more points to form clusters.

Does DBSCAN require choosing the number of clusters K beforehand?

No. DBSCAN discovers the number of clusters based on the density of the data.

Can a point initially labeled noise later join a cluster?

Yes. If the point is later found within epsilon distance of a core point, it can become a border point.

What happens when MinPts equals 1?

Every point becomes a core point because each point counts itself. Points connected within epsilon can form clusters.

What is the recommended rule of thumb for MinPts?

MinPts should be at least D + 1, where D is the number of dimensions. For noisy data, MinPts = 2D is suggested.

What is the purpose of a k-distance plot in DBSCAN?

To help choose an appropriate epsilon value by identifying the elbow in sorted nearest-neighbor distances.

Why can DBSCAN detect irregularly shaped clusters?

Because it groups points based on density connectivity rather than their distance to a centroid.

Why is DBSCAN considered robust to noise?

Because it can identify isolated points as noise instead of forcing every point into a cluster.

Why does DBSCAN struggle with varying densities?

Because it uses the same epsilon and MinPts settings throughout the dataset, which may not work well for clusters with different densities.

What is the goal of internal clustering evaluation?

To identify clusters whose points are close together while keeping different clusters well separated.

What does intra-cluster distance measure?

The distance between points belonging to the same cluster. Smaller distances are generally preferred.

What does inter-cluster distance measure?

The separation between different clusters. Larger distances are generally preferred.

What does the Davies-Bouldin Index measure?

How compact clusters are compared to the separation between their centers.

Is a higher or lower Davies-Bouldin Index better?

Lower is better because it indicates more compact and better-separated clusters.

Is a higher or lower Silhouette Coefficient better?

Higher is better. Values closer to +1 generally indicate better-separated and more cohesive clusters.