Is Mutual Information(MI) an important factor to consider in unsupervised learning (Clustering)?

Question

I have a supervised learning problem. The final step in the solving process is segmentation. Do features with the lowest MI affect the clustering process?

My problem about Churn Customers Segmentation: I found out some features with no MI at all. Do I drop these features?

Answer 1

You should do a feature important experiment, like this.

https://github.com/ash-wicus-ml/Notebooks/blob/master/XG%20Boost%20-%20Feature%20Importance.ipynb

When you know what your X-variable is, you can run some clustering exercises.

https://github.com/ash-wicus-ml/Notebooks/blob/master/Clustering%20Algorithms%20Compared.ipynb