Predicting customer churn with extended one-class support vector machine

As markets become increasingly saturated, customer churn prediction has become great concern to many industries. Class imbalance presents a particular challenge to customer churn prediction. To overcome this problem, this paper investigates the effectiveness of an extended one-class support vector machine approach to predict customer churn. The proposed model was compared with support vector data description, artificial neural network and decision tree. Result shows that the extended one-class support vector machine performs best among them in the aspect of hit rate, coverage rate, and lift coefficient.