Bisecting K-Means

Divisive hierarchical clustering that recursively splits clusters into two, combining aspects of hierarchical and K-Means clustering.

When to use:

• Want hierarchical structure with K-Means quality
• Need more consistent results than agglomerative
• Large datasets where regular hierarchical is too slow
• Want balance between speed and hierarchy

Strengths: More consistent than regular K-Means, faster than agglomerative hierarchical, creates hierarchy, good for text clustering Weaknesses: Still requires specifying k, slower than regular K-Means, assumes spherical clusters

Model Parameters

N Clusters (default: 8, required) Number of clusters to form.

Init Method (default: "random") How to initialize cluster centers:

• random: Random initialization (default for bisecting)
• k-means++: Smart initialization

Max Iterations (default: 300) Maximum iterations for each bisection.

Bisecting Strategy (default: "biggest_inertia") How to choose which cluster to split next:

• biggest_inertia: Split cluster with largest within-cluster variance (default)
• largest_cluster: Split largest cluster by number of points

Random State (default: 42) Seed for reproducibility.