Hybrid PPFCM-ANN model: an efficient system for customer churn prediction through probabilistic possibilistic fuzzy clustering and artificial neural network

Abstract

A vital issue for customer correlation management and consumer conservation in the telecommunications business is increased customer churn. The data mining approaches can aid in the prediction of churn behavior of consumers. This article aims to propose a system to predict customer churn through hybrid probabilistic possibilistic fuzzy C-means clustering (PPFCM) along with artificial neural network (PPFCM-ANN). This paper comprises of two modules: (1) proposing clustering component on the basis of PPFCM and (2) churn prediction component on the basis of ANN. The input dataset is gathered into clusters, with the help of probabilistic possibilistic fuzzy C-means clustering algorithm in the clustering module. The obtained clustered information is used in the artificial neural network, and this hybrid construction is further used in the churn prediction module. During the testing process, the clustered test data select the most accurate ANN classifier which corresponds to the closest cluster of the test data, according to minimum distance or similarity measures. Finally, to predict the churn customer the output score value is utilized. Three sets of experiments are carried out: the primary set of experiments comprises PPFCM clustering algorithm, the secondary set assesses the classification result, and the third set authenticates the proposed hybrid model presentation. The proposed hybrid PPFCM-ANN model provides maximum accuracy when compared to any single model.