DataTerminology.comNEW
A clustering method that builds a hierarchy of clusters through iterative merging (agglomerative) or splitting (divisive), visualized with a dendrogram.
DESCRIPTION
Hierarchical clustering produces a nested hierarchy of clusters without requiring a predefined number. Agglomerative methods start with singletons and merge; divisive methods start with one cluster and split. Linkage choices (single, complete, average, Ward) and distance metrics critically shape results. Dendrograms reveal structure and enable threshold-based cluster selection. It handles various cluster shapes but can be sensitive to noise and scaling. Complexity can be high for large datasets.
ALTERNATIVE DEFINITIONS
Wikipedia
Hierarchical clustering, also known as hierarchical cluster analysis, is a method of cluster analysis which seeks to build a hierarchy of clusters.
Defines the goal as building a hierarchy of clusters.
scikit-learn
Agglomerative clustering is a bottom-up hierarchical clustering method.
Emphasizes agglomerative strategy and hierarchy direction.
RELATED TERMS
Agglomerative Clustering Divisive Clustering Dendrogram Linkage Criteria Ward’s Method Single Linkage Complete Linkage Average Linkage Dissimilarity Measure
COMPETING TERMS
RELATIONS
- Broader: Clustering
- Narrower: Agglomerative Clustering
KNOWLEDGE AREAS
TARGET AUDIENCE
Data Scientist Data Analyst AI Engineer
EXTERNAL SOURCES
- Wikipedia - Hierarchical clustering
- scikit-learn - User guide for hierarchical methods
- SciPy - API for hierarchical clustering
- CRAN - Efficient hierarchical clustering
TOP QUESTIONS
How do I choose linkage in Hierarchical Clustering?
Single linkage favors chaining, complete linkage favors compact clusters, average balances, and Ward minimizes variance. Try several and validate with silhouette or domain knowledge.
Category: Best Practice
How many clusters should I cut from the dendrogram?
Use distance thresholds, elbow in linkage distances, or validation metrics. Consider domain interpretability over purely numerical criteria.
Category: Requirements
What is a common pitfall in Hierarchical Clustering?
Using unscaled features with Euclidean distance can distort results. Standardize or choose appropriate metrics for mixed data.
Category: Mistakes
Is Hierarchical Clustering suitable for large datasets?
Naively it’s O(n^2) memory/time. Use optimized implementations, approximate methods, or pre-aggregation for scalability.
Category: Performance
How does Hierarchical Clustering compare to K-means?
It does not require specifying k and can find nested structure, while K-means is faster but assumes spherical clusters and fixed k.
Category: Comparison
Can I use non-Euclidean distances?
Yes. Many implementations support cosine, correlation, or custom distances. Ensure linkage compatibility with the chosen metric.
Category: Technique
How to visualize results of Hierarchical Clustering?
Use dendrograms, heatmaps with clustered rows/columns, and cluster profiles summarizing feature means.
Category: Visualization
Does Hierarchical Clustering handle outliers well?
It can be sensitive. Consider robust distances, trimming, or pre-filtering outliers to prevent chain effects.
Category: Risk
What preprocessing improves Hierarchical Clustering?
Standardization, dimensionality reduction (PCA), and removal of redundant features generally improve stability and clarity.
Category: Best Practice