In today’s lecture, the professor discussed the instability of DBSCAN in comparison to K-means. The following scenarios illustrate DBSCAN’s instability:
Sensitivity to Density Variations:
DBSCAN’s stability is affected by variations in data point density. When density differs significantly across dataset segments, clusters with different sizes and shapes can form. Selecting appropriate parameters (e.g., maximum distance ε and minimum point thresholds) for defining clusters becomes challenging.
In contrast, K-means assumes spherical, uniformly sized clusters, making it more effective when clusters share similar densities and shapes.
Varying Cluster Shapes:
DBSCAN excels in accommodating clusters with arbitrary shapes and detecting clusters with irregular boundaries. This is in contrast to K-means, which assumes roughly spherical clusters, demonstrating greater stability when the data adheres to this assumption.