Have a look at our work
Customer churn modeling for a subscription-based services
To build a predictive model for a web service provider. The ultimate goal was to calculate Customer Lifetime Value in order to be able to predict future revenues by sales channels.
What we did
We developed a model which is predicting the probability of churn for 3, 6, 12 months periods. The following statistical methods were used: k-NN, Bayesian networks, Multinomial logit, XP boost.