AI FOR VET • MODULE 5

Discovering Hidden Patterns: K-Means Clustering

Grouping unlabeled telemetry without human supervision.

🎯 Learning Outcomes / Ishodi učenja:
  • Understand centroid initialization and Euclidean distance assignment.
  • Step through the iterative assign-update convergence loop.
  • Observe cluster inertia (Within-Cluster Sum of Squares) diminishing.
  • Identify under-clustering (K too small) vs over-clustering (K too large).

In unsupervised learning, data points have no labels. K-Means finds K cluster centers (centroids) by alternating between: 1) assigning points to the nearest centroid, and 2) recomputing centroids as the average of assigned points.

Inertia = ∑ min ||x_i − μ_j||²

Convergence occurs when centroids no longer move between iterations.

🔬 Interactive Laboratory / Interaktivna laboratorija:

🔍 Guided Investigation / Istraživački zadatak:
  1. Step-by-step loop: Click Step repeatedly and watch centroids jump to cluster centers.
  2. Wrong K: Select K = 2 on a 3-cluster dataset. Notice how two clusters are forced together.
  3. Random restarts: Click Reset with different seeds to see if initialization affects final groups.