Hierarchical Cluster Analysis

Hierarchical cluster analysis is an unsupervised method that builds a tree of nested clusters by repeatedly merging (or splitting) observations based on their pairwise distances.

계층적 군집분석은 관측치 간 거리를 바탕으로 군집을 반복적으로 병합(또는 분할)하여 중첩된 군집의 나무 구조를 만드는 비지도 학습 방법이다.

How do I group similar cases when I don't know how many clusters there are?

군집 수를 모를 때 비슷한 사례끼리 어떻게 묶나요?

Agglomerative hierarchical clustering starts with each case as its own cluster and successively merges the closest clusters using a linkage rule such as Ward's method, producing a dendrogram. It fits small to medium datasets where you want to inspect cluster structure at several levels before choosing the number of clusters.

When to use Hierarchical Cluster Analysis

• Grouping respondents, regions, or products on several numeric variables
• Exploring possible numbers of clusters by cutting a dendrogram
• Small to medium samples where a full distance matrix is manageable

When to use something else

• The dataset is very large and the number of clusters is roughly known: use k-means clustering
• Clusters should come from a probabilistic model with fit indices: use latent profile analysis or Gaussian mixture models

Assumptions and requirements

• Variables are on comparable scales (standardize if not)
• A meaningful distance measure and linkage method are chosen
• Results are sensitive to outliers and to the linkage choice
• Memory grows with the square of the number of cases

What the analysis reports

• Dendrogram
• Agglomeration schedule (merge heights)
• Cluster membership for the chosen number of clusters
• Cluster profiles (means per variable)
• Validity indices such as silhouette width

How to report it

Hierarchical cluster analysis using Ward's method and squared Euclidean distance identified [k] clusters (average silhouette width = [value]).

Related methods

• K-Means Clustering
• Latent Profile Analysis (LPA)
• DBSCAN

Run Hierarchical Cluster Analysis on Quorant