BIRCH

Balanced Iterative Reducing and Clustering using Hierarchies - memory-efficient hierarchical clustering for very large datasets.

When to use:

• Very large datasets that don't fit in memory
• Need fast online clustering
• Roughly spherical clusters
• Memory constraints

Strengths: Very fast, memory efficient, online learning, handles large datasets, incremental Weaknesses: Assumes spherical clusters, sensitive to threshold, order-dependent results

Model Parameters

N Clusters (default: 3, required) Number of clusters after final clustering step.

Threshold (default: 0.5) Radius of the subcluster obtained by merging a new sample. Key parameter.

• Low (0.1-0.3): Many small clusters, high memory
• Medium (0.5): Balanced
• High (1.0+): Few large clusters, low memory

Branching Factor (default: 50) Maximum number of subclusters in each node.

• 10-30: Deeper tree, slower
• 50: Good default
• 100+: Shallower tree, faster