Home / Summaries / Marketing Models / cluster-customer-clusters

Market segmentation and cluster analysis - Clustering algorithms - Hierarchical clustering models

Q: What are divisive techniques?

All customers begin in one segment and each iteration breaks off the customer, customers or cluster segment that is the most different . Continues until everyone is in his or her own cluster .

Q: What is single-link clustering?

It puts a customer into a segment of he or she is similar enough to at least 1 member in the cluster. If the customer is in the middle of 2 clusters, the algorithm looks at who is the closest in either of the clusters and joins the customer there.

Q: What is complete-link clustering?

Customers join a cluster only if they are similar to all the other members . For the customer in the middle the algorithm will look at the ones most different from the customer in each cluster and joins it in the other cluster .

Q: What is average-link clustering?

Looking at averages to place a customer with a cluster. Thus taking the average value of X and Y of a cluster and plotting this ( cluster centroid ). Customer is then placed in the cluster of which the cluster centroid is the closest .

Q: What is a dendrogram?

A dendrogram is a tree-like structure that shows the steps during which each car class combined with the others. Numbers at each step are fusion coefficients .

Q: How do you know if minimum variance is achieved?

With R^2. R^2=(SStotal-SSerror)/SStotal. Total sum of squares measures how close every customer is to the overall means on all the variables used in the analysis. Error sum of squares is computed by comparing each individual customer to the means of only the other observations in the customer's cluster .

Q: How does Ward's method work?

Start with each of the N sample units and calculate SStotal . SSerror is the same as SStotal because there are no clusters yet. Begin forming clusters of size 2 and leave the other customers in their own cluster (N=1). Then have N-1 clusters . Do this until all options have been done . The pair that forms the smallest SSerror or the largest R^2 forms the first cluster . Then form N-2 clusters etc . Until all sample units are combined into a single large cluster of size N.

7 important questions on Market segmentation and cluster analysis - Clustering algorithms - Hierarchical clustering models

What are divisive techniques?

All customers begin in one segment and each iteration breaks off the customer, customers or cluster segment that is the most different. Continues until everyone is in his or her own cluster.

What is single-link clustering?

It puts a customer into a segment of he or she is similar enough to at least 1 member in the cluster. If the customer is in the middle of 2 clusters, the algorithm looks at who is the closest in either of the clusters and joins the customer there.

What is complete-link clustering?

Customers join a cluster only if they are similar to all the other members. For the customer in the middle the algorithm will look at the ones most different from the customer in each cluster and joins it in the other cluster.

What is average-link clustering?

Looking at averages to place a customer with a cluster. Thus taking the average value of X and Y of a cluster and plotting this (cluster centroid). Customer is then placed in the cluster of which the cluster centroid is the closest.

What is a dendrogram?

A dendrogram is a tree-like structure that shows the steps during which each car class combined with the others. Numbers at each step are fusion coefficients.

How do you know if minimum variance is achieved?

With R^2. R^2=(SStotal-SSerror)/SStotal.
Total sum of squares measures how close every customer is to the overall means on all the variables used in the analysis.
Error sum of squares is computed by comparing each individual customer to the means of only the other observations in the customer's cluster.

How does Ward's method work?

Start with each of the N sample units and calculate SStotal. SSerror is the same as SStotal because there are no clusters yet.
Begin forming clusters of size 2 and leave the other customers in their own cluster (N=1). Then have N-1 clusters.
Do this until all options have been done.
The pair that forms the smallest SSerror or the largest R^2 forms the first cluster.
Then form N-2 clusters etc.
Until all sample units are combined into a single large cluster of size N.