Prueba de agrupación en clústeres de DBSCAN

Las respuestas y explicaciones aparecen a medida que realiza el cuestionario.

1. Según la convención de scikit-learn, ¿qué califica un punto como punto central?
   • Its epsilon neighborhood contains at least min_samples, including itself
   • It is the closest point to a centroid
   • It has no neighbors within epsilon
   • It lies outside every dense region
2. ¿Cómo puede un punto fronterizo pertenecer a un cluster sin ser núcleo?
   • It has more neighbors than every core point
   • It is reachable from a core point but lacks enough neighbors itself
   • It is assigned to the nearest centroid
   • It has a zero distance to every sample
3. ¿Qué significa una etiqueta de ruido DBSCAN?
   • The point cannot belong to a cluster under any settings
   • The record is definitely erroneous
   • The point was not assigned under the chosen metric and parameters
   • The prediction probability is zero
4. ¿Qué puede suceder cuando épsilon se establece en un valor demasiado grande?
   • A too-large radius can split one connected dense region into separate groups
   • Increasing the radius can make previously core points non-core
   • Epsilon no longer affects which observations are neighbors
   • Nearby dense groups may merge
5. ¿Por qué las unidades de características inconsistentes pueden distorsionar los resultados de DBSCAN?
   • They change the number of neighbors required for a point to be core
   • A large-scale feature can dominate distance calculations
   • They make epsilon independent of measurement units
   • They affect cluster labels only after the neighborhood graph is fixed
6. A compact cluster is found while a diffuse group becomes noise under one epsilon. What limitation is illustrated?
   • Core points cannot connect
   • DBSCAN requires a predetermined cluster count
   • One global density threshold may not suit clusters with different densities
   • The metric no longer affects neighborhoods
7. How does DBSCAN differ from k-means regarding cluster count?
   • Both require a count known from labels
   • DBSCAN always creates exactly two clusters
   • K-means selects arbitrary-shaped regions without centroids
   • DBSCAN does not require the number of clusters as an input
8. Why is assigning future points to DBSCAN clusters not automatic in the basic procedure?
   • The method clusters the fitted sample and has no inherent new-point rule in its basic form
   • New points are assigned to the nearest cluster centroid by definition
   • The noise label is a calibrated probability that a point is an outlier
   • Every new point is assigned to the most common existing cluster