|
Abstract
|
Customer churn is the main concern of most service-based firms such as telecommunication operators. Customer churn
prediction is a useful tool to predict customers who are at risk of churning. Many prediction models such as data mining
models are presented to detect churning customers. Neuro-fuzzy models have been successfully used in many different fields
resulting in very accurate prediction. However, the performance of these techniques was not completely explored in churn
prediction setting. In this paper, Adaptive Neuro-Fuzzy Inference System (ANFIS) and Locally Linear Neuro-Fuzzy (LLNF) with
Locally Linear Model Tree (LoLiMoT) learning algorithm are applied to three telecommunication datasets. These Neuro-Fuzzy
classifiers are benchmarked to two of the most widely used neural network architecture such as the Multi Layer Perceptron
(MLP) and Radial Basis Function (RBF). The results of the benchmarking experiments show that in most of the cases (churn
datasets) these Neuro-fuzzy techniques exhibit a high performance and perform better than constructed neural network
models of MLP and RBF.
|