We have this ‘Telco Churn’ data, which has information about each customer of this telco company along with whether if a given customer has chunred or not.
I wanted to understand what makes customers churn or not, and also wanted to know what we could do to prevent the churn.
First, I have run Random Forest and Logistic Regression by select all the variables except Total Charge, which I saw was highly correlated to Tenure.
From the result of both analysis with Random Forest and Logistic Regression, I found that Tenure has a high correlation with Churn.
I confirmed this with a density plot like the below.
So, I have decided to use Tenure as a proxy to Churn and position this problem as a Regression problem rather than Classification problem.
Therefore, I have decided to use Linear Regression and Random Forest to see what makes Tenure longer or shorter.
By the way, this might make sense better because at the end of the day what we want is to have customers stay longer on the service, which usually will result in more revenue.
Anyway, here’s the result with Linear Regression.
And here’s one with Random Forest.
We can see that Contract increases the length of Tenure.
By looking at the effect of each variable with Random Forest, ’Two Year" contract would increase the length of Tenure. This is consistent with what we can see from the Linear Regression’s coefficient chart above.
And finally, here is a density chart that shows the distribution of Tenure colored by Contract Type (e.g. Two Year, etc.), and separated for Churned customers and for Not Churned customers.
It’s hard to see the difference between the Churned and the Not Churned, but we can clearly see the difference between Two Year and Month-to-month contracts.
Among other variables that might have effects on whether the customer churn or not, I would suggest that we will focus on our effort on converting the customers’ cotract to One or Two years contract.