* I plotted feature importance for a simple Randomforest Classifier without any feature engineering, then after seeing the important features, i decided to create more features by multiplying, suming etc some of them. while some where just by intuition.

* I didn't know until i tried building a model with and without both, then i noticed improvements. Creating features based on TENURE and REGION can also give the model a sense of customer behavior with respect to how long they've been with the service provider or the region where they are based

* I tried the usual statistical methods but this one seem to perform better* I tried the usual statistical methods but this one seem to perform better

Oh! That's cool, thanks