the.com/elbow method
you bend the graph until it bends back, and that's your answer.
means a way to pick the optimal number of clusters in k-means by plotting variance explained against cluster count and eyeballing where adding more clusters stops helping much.
from emerged alongside k-means clustering in mid-20th century statistics; the name is pure visual pun, the within-cluster variance curve drops steeply then flattens, tracing the silhouette of a bent arm.
the metricwithin-cluster sum of squares, wcss, plotted per k
the flawthe bend is often ambiguous or missing entirely
the rivalsilhouette score gives a less subjective second opinion
the shapecurve always decreases, never increases, by construction
for instance
customer segmentation — retailers pick 4-6 clusters this way for marketing personas
image compression — k-means color quantization uses it to choose palette size
scikit-learn tutorials — the canonical teaching example for unsupervised learning courses