Partition cost-sensitive CART based on customer value for Telecom customer churn prediction

Telecom Customer churn prediction is a cost sensitive classification problem. Most of studies regard it as a general classification problem use traditional methods, that the two types of misclassification cost are equal. And, in aspect of cost sensitive classification, there are some researches focused on static cost sensitive situation. In fact, customer value of each customer is different, so misclassification cost of each sample is different. For this problem, we propose the partition cost-sensitive CART model in this paper. According to the experiment based on the real data, it is showed that the method not only obtains a good classification performance, but also reduces the total misclassification costs effectively.