K-Means Clustering | Interactive Workspace

Watch K-Means Clustering learn

Follow each update in real time. Scrub or play to inspect the algorithm.

Step 1 of 9

02 Interactive demo

CURRENT LOSS · inertia

—

Objective during training: inertia (WCSS)

WHAT'S HAPPENING

Unlabeled Data

We start with unlabelled observations scattered in feature space. Without any target labels, K-Means will discover K = 3 natural clusters purely by analyzing the spatial distances between points.

ELBOW METHOD · MODEL SELECTION

Inertia (WCSS) vs Number of Clusters (K)

Active: K = 3

As the $K$ value increases, WCSS (inertia) decreases because adding more clusters reduces the within-cluster variance. However, beyond a certain point, the marginal gain diminishes, and the plot exhibits an elbow shape. The optimal $K$ is chosen at the elbow, where the reduction in WCSS becomes insignificant.

Switch K:

Optimal Elbow (K = 3)Inertia: 8,081

Sweet spot: Every natural blob gets its own dedicated centroid. Inertia drops precipitously by ~72% down to 8,081. This is the optimal elbow point!