Hierarchical
clustering methods
Distance optimizing combinatorial/agglomerative
procedures
- Single link (nearest neighbor)
- Complete link (farthest neighbor)
- Average link (UPGMA)
- Single link (WPGMA)
- Centroid
- Median
- beta-Flexible
- Flexible UPGMA
Homogeneity-optimizing
combinatorial/agglomerative methods
- Incremental sum of squares (Ward)
- Min. sum of squares in new clusters
- Incremental variance
- Minimum variance in new clusters
- Minimum average distance in new clusters
- Minimum increase of average distance
- lambda-Flexible
Information theory clustering
- Min. pooled entropy in new clusters
- Min. increase of pooled entropy
- Min. mutual information of vars in new clusters
- Min. increase of mutual information
- Division by pooled entropy
- Division by mutual information
Miscellaneous techniques
- Global optimization
- Ordinal clustering
- Minimum spanning trees
- Neighbor joining (unrooted, midpoint rooted, outgroup rooted)
Non-hierarchical
clustering (partitioning)
k-means clustering
- Initialization from random or input partitions, random or input
seed objects or from farthest points.
Multiple partitioning
Quick clustering
- About 15 coefficients used.
Global optimization
- More than 30 coefficients used.
- Initialization from random or input partitions, random or input
seed objects or from farthest points.
Ordinal clustering
- Best suited to ordinal coefficients.
Fuzzy c-means clustering
- Ternary plots showing classification for three clusters.
- Best method for finding less separated clusters.
Ordination
methods and their options
Principal components analysis (PCA)
- Standardized, centered, non-centered variants.
- Rohlf, Euclidean and Mahalanobis biplots.
Principal coordinates analysis (PCoA) and
Nonmetric Multidimensional Scaling (NMDS)
- More than 30 coefficients employed.
Correspondence analysis (CoA)
- Symmetric weighting, variables at barycenter of objects or vice
versa in joint plots.
Canonical correspondence analysis (CCoA)
- WA scores: Symmetric weighting, variables at barycenter of objects
or vice versa.
- LC scores.
Canonical correlation analysis (COR)
Rendundancy analysis (RDA)
Canonical variates analysis (CVA)
- Spherized vs non-spherized scores.