A new filter feature selection approach for customer churn prediction in telecommunications

There is little literature to introduce the approaches for the feature selection, which plays an important role in the customer churn prediction. In addition, due to the imbalanced data classification problem occurring, most of the traditional approaches ineffectively select the important features for the churn prediction. This paper proposes a new filter feature selection approach for customer churn prediction in telecommunications. The main idea of this approach is to calculate the dependency between each input feature and the class. Finally, the comparative experiments were carried out, and the results show that the new proposed feature selection approach is very effective for the churn prediction.