DBSCAN — Density-Based Clustering
K-Means and agglomerative clustering both struggle with one fundamental shape: non-convex clusters. Real data often has crescent shapes, ring structures, or elongated filaments that no centroid-based method can capture correctly. DBSCAN (Density-Based Spatial Clustering of Applications with Noise) solves this by defining clusters as dense regions of points separated by sparse regions. It requires no K upfront, automatically identifies noise points (outliers), and finds clusters of arbitrary shape — making it a powerful complement to the methods you have already learned.
Analogy🏏Cricket
🏏 Think of it like cricket: picture a stadium with supporters clustered in stands. In each stand, fans are packed densely — K-Means would try to draw a circle around each stand, but the shapes are irregular (an arc, a corner block, a gallery tier). DBSCAN instead asks: 'Is this fan surrounded by enough other fans within a small radius?' If yes, they are a core fan of that stand. Dense fans pull nearby fans into their cluster. Isolated fans who wandered in alone — perhaps a journalist or a scout sitting outside the stands — are marked as noise. The shape of each stand can be any polygon; DBSCAN doesn't care.