Our example Scenario
The title of this post uses the word “experiment” since I tried experimenting, leveraging cluster analysis for supplier selection. I am not sure if this approach is used at a production scale anywhere yet as far as supplier selection goes.
Let us create a scenario that we will then use to walk through the cluster analysis process
The pre-qualification criteria for Supplier ranking that the procurement manager has in mind are as shown in the illustration below:
Supplier criteria values
Generic overview of Cluster Analysis
Clustering process begins with formulating the problem and concludes with carrying out analysis to verify the accuracy and appropriateness of the method. The clustering process has the following steps:
- Formulate the problem and identify the selection criteria
- Decide on the number of clusters
- Select a clustering procedure
- Plot the Dendogram ( A Dendogram is a tree diagram used to illustrate the output of clustering analysis) and carry out analysis to compare the mean across various clusters
Leveraging Cluster Analysis on our scenario
In the first step, every supplier is rated on a scale of 0-1 for each attribute as shown in the table below:
In this step we need to decide on the number of clusters. We want the initial list of suppliers to be split into two categories, good suppliers (shortlisted) and bad suppliers (rejected) hence our number of clusters is two.
Next, we apply both hierarchical and partitional clustering methods to supplier data. We choose the method which has the highest R Square value pooled over all the 14 attributes. A summary table showing those pooled values is shown below:
The R Square value for k-means is the highest among different methods, hence k-means is chosen for clustering. There are several other methods available for determining the goodness of fit.
In this step, we apply the k-means clustering to the given data. Graphical comparison of the two clusters is shown in the figure below. This graph helps us identify the cluster which has the suppliers of higher mean scores. Cluster 1 has 6 members and cluster 2 has 14 members. From the figure below it can be seen that suppliers of cluster 1 has better mean scores than suppliers in cluster 2 (on most criteria); hence 6 suppliers of cluster 1 are chosen as shortlisted suppliers.