I have a data of customers and i want to split them to segment (profiling). The columns of the data are Amount-Spending, Amount-Bonus, Age, Churn-or-Not. So i clustered my data with k-means. to 3 clusters. Now, how can i show that the first column of the data ('Amount-Spending') is more close to cluster number 1 for example. So i can say that cluster number 1 are people that usually spend a lot of money and not just, customer number 5 is in cluster 1.

  • $\begingroup$ You can try and measure correllation. $\endgroup$ – mathreadler Oct 6 '15 at 10:33

It seems to me that you are in danger to misuse Cluster Analysis, which is a technique to discover structure in data and not to describe a structure that one just assumes to exist. To see what I mean, ask yourself the following questions for example:

  1. Why are you clustering into 3 groups? Not 4 or 9?
  2. When plotting the data (luckily there are 3 attributes only), do you see any groups? How many? How can these groups roughly be described using the values of the attributes?
  3. KMeans uses the Euklidean distance as a measure of similarity of samples. Is this the appropriate measure in your case?
  • $\begingroup$ Very good arguments. +1 $\endgroup$ – mathreadler Oct 6 '15 at 10:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.