Spatial analysis · 4 min
Spot clusters on a map
When dots bunch together, how to find the groups properly instead of by eye.
Spatial analysis
Your eyes are a poor judge
People see groups in any scatter of dots, even random ones. Zoom out and two clusters merge; zoom in and one splits.
A clustering method uses a fixed rule instead, so the answer doesn't change with the zoom level or who is looking.
Lesson notes
Everything from the cards above, on one page to come back to.
Your eyes are a poor judge
People see groups in any scatter of dots, even random ones. Zoom out and two clusters merge; zoom in and one splits.
A clustering method uses a fixed rule instead, so the answer doesn't change with the zoom level or who is looking.
Distance-based: who has neighbours?
One common rule, called DBSCAN: pick a distance and a minimum number of points. A point with at least that many points within that distance, itself included, is the core of a cluster. Points close to a core join its cluster.
Points that join nothing are noise: lone points with no crowd around them.
Distance 500 m, minimum 5 points: a café with four others within 500 m is a core point.
K-means: you choose how many
K-means works the other way round: you say how many groups you want, and every point is put in the group whose centre is nearest. There is no noise; every point gets a group.
It suits questions like “split our customers into 4 delivery regions”. It suits “where are the real hot spots?” less well.
In MapGO
Find clusters, in the Process tools, offers both methods. It writes a cluster_id column numbering each group and, for distance-based runs, a cluster_type column: core, edge or noise. It also colours the points by cluster.
Questions
What is the difference between DBSCAN and k-means?
DBSCAN groups points that have enough neighbours within a distance and leaves lone points as noise. K-means splits every point into a number of groups you choose.
What does noise mean in clustering?
Points that belong to no cluster, because they don't have enough neighbours close by.