The Edge of a Nearest‑Centroid Cluster
The nearest‑centroid rule defines cluster membership by which centroid is closest, so cluster edges are the straight perpendicular bisectors between centroids — independent of the actual spread, shape, or density of the points.
A complete interactive classroom, not just a preview.
Start when you are ready to enter this Stage's 8 scenes and explore, respond, and learn as you go.
What does the nearest‑centroid rule really tell us about a cluster's edge?
Most diagrams show clusters as round blobs, but real cluster edges are where the nearest‑centroid rule quietly makes a decision.
Intuition says a point belongs to a cluster if it is 'close' to it — but 'close to what?' The centroid, the nearest point, or something else entirely?
Compare a nearest‑centroid boundary against a nearest‑neighbor boundary on the same 2‑D point cloud, so the edge geometry becomes visible.
The nearest‑centroid rule carves space with straight‑line bisectors between centroids, which is exactly why those edges look so different from the data.
A cluster's edge should follow the outer points of the group, hugging the real shape of the data.
- k‑means optimization and convergence
- Choosing the number of clusters k
- Distance metrics beyond Euclidean
- Probabilistic or soft clustering
- 01Where does a cluster end?slideQuestion
Open with a small 2‑D scatter of two point clouds and ask the learner where the boundary between them actually lies.
- Clusters are usually drawn as round blobs
- The rule says: assign each point to its nearest centroid
- So the edge is a geometric object, not a data outline
- 02Predict the edgequizPrediction
Learner commits to one mental model of the cluster edge before the geometry is revealed.
- Choose between: a curve hugging the points, a perpendicular bisector between centroids, or a circle around each centroid
- 03A picture of the rule in actionslideEvidence
Show two clouds with their centroids marked, then color every point by its assigned centroid. The colored regions meet along a visibly straight line, even where the data bulges.
- Each point is assigned to the closer of the two centroids
- The boundary runs straight through the space
- Points that 'look like' they belong to one cloud can be assigned to the other
- 04Drag the centroids, watch the edge moveinteractiveEvidence
Learner drags two centroids over a fixed point cloud and sees the straight bisector redraw in real time, while the assigned color of each point updates instantly.
- The edge is always a straight line
- It always sits exactly halfway between the two centroids
- Points change sides the moment a centroid crosses the midpoint to them
- 05Why the edge is a bisectorslideExplanation
Derive the boundary algebraically: a point x is assigned to centroid c1 when ||x − c1|| < ||x − c2||. Squaring both sides and canceling x·x collapses the inequality to a linear equation in x, whose solution set is a line.
- ||x − c1||² < ||x − c2||² simplifies to a linear inequality
- The set where both sides are equal is a perpendicular bisector
- The line is perpendicular to c1c2 and passes through the midpoint
- 06When the picture liesslideBoundary
Show an elongated, unequal cluster paired with a tight round one. The nearest‑centroid bisector can place almost the entire elongated group inside the tight cluster's region.
- The rule ignores cluster shape and spread
- Equal distance to the centroid does not mean equal distance to the data
- This is why nearest centroid is a weak default for irregular clusters
- 07Three centroids, three straight wallsslideTransfer
Extend the same rule to three centroids: the space splits into three straight‑edged regions (a Voronoi diagram). The walls never curve to follow the points.
- Each pair of centroids contributes one straight wall
- Three walls meet at a single equidistant point
- Walls depend only on centroid positions, not on the data
- 08What the rule really says about the edgeslideResolution
Close the loop: state the answer in one sentence and contrast it with the original intuition.
- The edge is a perpendicular bisector between centroids, not a contour around the data
- It is a property of the centroids, not of the points
- That is why nearest‑centroid boundaries look so 'unnatural' on messy data
Discussion threads for a Stage aren't available yet.
This path ends here.
