RapidMiner Tutorial: K-means Clustering

What is K-means?

en.proft.me

A unsupervised correlational technique that groups together like types of observations in a data set.

The "K" in K-means clustering implies the number of clusters the user is interested in. In other words, the user has the option to set the number of clusters he wants the algorithm to produce.

What data are we going to use?

We're going to use a made-up data set that details the lists the applicants and their attributes. The attribute names are all self-explanatory. It has 546 rows and 4 columns.

Goal

We would like to cluster applicants using their physical and mental attributes in order to make it easier to pick the right people.

TL;DR

RapidMiner Process

1. Look for the Read CSV operator in the Operator panel
2. Drag it to the Process panel 
3. Connect the operator's Out to Res
4. Click on Import Configuration Wizard in the Parameters panel
   
5. Look for our example file, click it, and then click Next
6. In the Data Wizard window, make sure you pick Comma in the Column Separation section since we are using comma separated values (.CSV) type of file then click Next twice
   
7. Make sure the data types are all correct and then click Finish
   
8. Look for the Select Attributes operator and 
9. Drag it to the Process panel
10. In the Parameters panel, click on Attribute Filter type dropdown menu and pick Single so that we can isolate only one attribute in the data set
11. In the Attribute dropdown menu, pick Training Course to tell RM that it's the attribute we want to isolate
12. Click on the Invert Selection checkbox to indicate that we want to exclude Training Course in the actual calculation
    
     
13. Next, look for the Set Role operator and drag it to the right of Select Attributes
14. Connect the operators via their Exa nodes
15. In the Parameters panel, click on the Attributes Name dropdown menu and pick Applicant
16. Click the Target Role dropdown menu and pick ID. This will make the Applicant variable an identifier of all our observations so we do not just get anonymous results
    
17. Look for the Normalize operator and drag it to the right of the Set Role operator. Connect them via their Exa nodes. We do not need to modify any of the parameters for now. Normalization is used so that no particular attribute will over-influence the clustering
18. Look for the K-means operator and drag it to the right of the Normalize operator. Connect the two via their Exa ports. Connect the K-means operator's Clu port to the Process panel's Res port to the right
    
19. Press F11 on your keyboard to Run Process. Once it is done, RapidMiner will automatically switch to Results View
20. You can look at the Cluster Model Description to find out the number of observations per cluster
    
21. You can check the names of the people per cluster in the Folder View. This is the reason why we designated the ID role to Applicant using the Set Role operator earlier
    
22. You can also check the means of the centroids per cluster in the Centroid Table