The "churn" data set was developed to predict telecom customer churn based on information about their account. The data files state that the data are "artificial based on claims similar to real world." The data consist of 19 predictors related to the customer account, such as the number of customer service calls, the area code, and the number of minutes. The outcome is whether the customer churned.

The data are contained in the C50 package and can be loaded using:

> library(C50)

> data(churn)

> ## Two objects are loaded: churnTrain and churnTest > str(churnTrain)

> table(churnTrain$Class)

a) Explore the data by visualizing the relationship between the predictors and the outcome. Are there important features of the predictor data themselves, such as between-predictor correlations or degenerate distributions? Can functions of more than one predictor be used to model the data more effectively?

b) Fit some basic models to the training set and tune them via resampling. What criteria should be used to evaluate the effectiveness of the models?

c) Use lift charts to compare models. If you wanted to identify 80% of the churning customers, how many other customers would also be identified?

Show your R code for each question.