K-Means

• Clustering algorithm
• Number of clusters is a hyper-parameter

Algorithm

• Start with K random centroids (points in ${\mathbb{R}}^p$)
• Repeat until convergence:
  • Assign each point of the data to its closest centroid
  • Replace current centroids with the mean of the points in the corresponding cluster

source: https://www.maartengrootendorst.com/assets/images/posts/2019-07-30-customer/kmeans.gif

Pros

• Easy to implement
• Easy to interpret

Cons

• Susceptible to correlated features
• Susceptible to features with different variances

See also

• LDA (Linear Discriminant Analysis) for supervised clustering
• PCA (Principal Component Analysis) to decorrelate and normalize features