Minimum Points (MinPts)

A useful way to think about Minimum Points (MinPts) is this: when does a small crowd become a real gathering instead of just a few people standing near each other? In density-based clustering, MinPts is the rule that answers that question. It sets the minimum number of nearby data points required for a region to count as genuinely dense.

In algorithms like DBSCAN, a point becomes a core point only if at least MinPts points lie within its neighborhood radius, usually called epsilon (ε). Those core points form the backbone of clusters. Points close to a core point but not dense enough themselves are border points, and isolated points are labeled noise. So MinPts works together with ε: ε defines “nearby,” and MinPts defines “dense enough.”

This parameter strongly affects what the algorithm sees as structure:

• If MinPts is too low, random small groupings can be mistaken for clusters.
• If MinPts is too high, meaningful sparse clusters can disappear and get labeled as noise.
• Higher MinPts usually makes DBSCAN more conservative and more resistant to outliers.

In customer segmentation, a low MinPts might create fake micro-segments from a handful of unusual shoppers. In transaction anomaly detection, a higher MinPts helps ensure that only truly dense spending patterns become clusters, leaving suspicious isolated behavior outside them.

A common practical guideline is setting MinPts to at least dimension + 1, with larger values for noisy data. In scikit-learn, this appears as the min_samples parameter in sklearn.cluster.DBSCAN. People often inspect a k-distance plot to tune ε, while choosing MinPts based on data dimensionality, expected cluster thickness, and noise level. MinPts matters because it controls the line between meaningful density and accidental proximity—the core idea behind density-based clustering.