Unsupervised Learning: K-means

Centroid

Non-data point that indicates the center of cluster as identified by K-means.

Operation:

1. Deploy N centroids randomly (N proportional to the number of expected classes).

2. Assigns, randomly, data points to classes.

3. Repeat iteratively.

   1. Compute the center of gravity of each class;

   2. Centroid is repositioned in that center of gravity;

   3. Update boundary.

4. Stop when updates become negligible.