Mean Shift

Centroid-based algorithm that automatically discovers clusters by finding density peaks in the feature space.

When to use:

• Don't know number of clusters in advance
• Clusters have different shapes
• Want to find natural cluster centers
• Have moderate-sized datasets

Strengths: Automatically finds number of clusters, handles arbitrary shapes, single parameter (bandwidth) Weaknesses: Very slow on large datasets, sensitive to bandwidth parameter, computationally expensive

Model Parameters

Bandwidth (optional) Size of the search window. Critical parameter.

• null: Automatically estimated from data (recommended)
• Low values: Many small clusters
• High values: Few large clusters
• Use estimate_bandwidth() for data-driven selection

Bin Seeding (default: false) If true, initial kernel locations are discretized to a grid for speed.

• false: Seed from all points (slower, more accurate)
• true: Seed from grid (faster, approximate)

Cluster All (default: true) Whether to cluster all points or leave orphans as outliers.

• true: Assign all points to nearest cluster
• false: Points far from cluster centers remain unassigned

Max Iterations (default: 300) Maximum iterations per seed point before convergence.