Spectral Clustering

Spectral Clustering uses a similarity graph and its mathematical properties (eigenvalues) to discover complex, non-convex cluster shapes that traditional algorithms like K-Means cannot handle.

It is particularly effective when clusters have irregular, non-spherical boundaries.

Spectral Clustering excels at finding clusters with non-convex shapes (e.g., L-shaped, crescent-shaped regions). If K-Means fails to capture the natural grouping in your data, try Spectral Clustering.

Use Cases

• Complex spatial patterns: When building layouts or urban patterns form non-convex shapes (e.g., L-shaped building clusters, ring-shaped arrangements).
• Connectivity-based grouping: Group geometries based on their adjacency relationships rather than just distance.
• Irregular cluster shapes: When K-Means produces unsatisfactory results because clusters are not spherical.