Hi Aarthi, from what I can see, your clustering seems to be holding up pretty well between the training and validation sets. That ±10% variability you're seeing is pretty good. In the real world, you're always going to have some differences when you apply your model to new data. The trends look fairly consistent across the board, which is a good sign. Cluster 2 seems to be your low-risk group, while Clusters 1 and 3 are picking up on higher-risk individuals. The fact that this pattern shows up in both datasets is promising.

As for best practices, you're already on the right track by comparing your training and validation results like this. One thing you might want to try is k-fold cross-validation to really give your clustering a challenge. You could also look into measures like the Rand index or Jaccard coefficient to quantify the similarity between your clusterings. And don't forget about visualizing your clusters - sometimes a good scatterplot can reveal things that numbers alone miss. The key is to keep testing your clustering on different subsets of your data and maybe even on some completely new data if you can get your hands on it. If it keeps holding up, you'll know you've got something solid.

Cheers!

Thank you Tom Duenas so much for your thoughtful and encouraging response. I really appreciate your insights and the time you took to help me.

Aarthi Narayan Hi Aarthi, what do the percentages for each indicator in each cluster mean, and how did you generate the percentages? I’m studying cluster analysis these days and trying to figure out whether we can quantify the contribution or importance of each indicator in each cluster. I would really appreciate it if you could discuss the details.