Abstract In this paper, on account of the complexity of customer churn in communication industry, fusing the advantages of rough sets, neural network and artificial bee colony algorithm(ABC), a new customer churn prediction model is put forward, which is a linearfused multiple classifier based on rough sets theory, neural network and artificial bee colony algorithm. Firstly, it completes the unsupervised separation of the continuous attributes using SOM; secondly, it reduces the discrete attributes using rough sets theory; thirdly, it builds four subclassifiers on the reduced attribute set using BP neural network, radial basis function neural network (RBF), ELMAN neural network and generalized regression neural network (GRNN); finally, it integrates linearly the prediction results from the subclassifiers and optimize the weights by ABC. Through applying the model to customer churn research in a telecommunication enterprise, the experiments results suggest that the integration technique is feasible and very efficient.
|