1. Consider all data points as one cluster
2. Identify and split the least similar observations
3. Repeat splitting until all clusters are divided

link: [towardsdatascience.com/understanding-the-concept-of-hierarchical-clustering-technique-c6e8243758ec](https://towardsdatascience.com/understanding-the-concept-of-hierarchical-clustering-technique-c6e8243758ec) 

Medium: Divisive approach algorithm

- Divisive Clustering takes into account the global distribution of data and so forms more accurate clusters ( i.e. doesn't get thrown off by local minima/maxima ) than Agglomerative Clustering.
- Agglomerative Clustering has O($n^2$) complexity in the most efficient implementation of the algorithm. Divisive Clustering is linear in the number of clusters.

Divisive vs. Agglomerative Clustering Algorithms

Start with all observations in one cluster, then splits form as one moves down the hierarchy.

University of Michigan - Ann Arbor

1. Divisive: Top-down approach
2. Agglomerative: Bottom down approach


Types of hierarchical clustering

Divisive: Top-down approach

One cluster for each observation, then clusters are moved together as one moves up the hierarchy of clusters

Steps: 
1. Each data point is initially in its own cluster with one item 
2. Most similar cluster of each stage is merged 
3. Stopping measure: number of clusters 

Linkage criteria
- Ward's method
- least increase in total variance (around cluster centroids) 
2. Average linkage 
- average distance between clusters
3. Complete linkage 
- max distance between clusters




Learn Before

Related

Learn After