Will try to find out causation for churn especially that could be connected with action.
Before starting analysis, I will look into the data at first. I found that tenure might be a kind of the answer to predict churn, because tenure gets longer, churn rate decreases. However, we cannot do with the tenure so I will proceed excluding this variable.
And for next, found that some variables have nothig with the customer who does not have internet service. So I will split analysis those who has internet servince and who does not.
On top of that, as nature, from perspective of churn, it could be said that this data is inbalanced data which means we should consider doing smote later.
To check importance for variables, I will do Boruta without smote at first. Now I can see F score does not showing up good socre.
So I will run Boruta with smote next. Now it seems F score gets better.
Seems threre are a lot of confirmed variables so will go to Logistic regression next, to see if we can find some causality.
Will run logistic regression by marginal effect doing smote
Seems internet service itself is effecting churn so I will repeat this and make several models
It might be nature, but having longer contract is making less churn for every group so we can give more advantage to have longer contract.
For user who has internet service, electronic check for paymenet method is accelerating churn from mailed check. May be we can have a switch campaign
For DSL users, service like Tech support and online security is making less churn so we can promote these service more. In addition, use of streaming movies are making much churn. May be this is giving customer bad experience so may be there are place to improve their services.