Hybrid models and error weighting for predicting customer churn in telecom industry

Abstract

It is more expensive to acquire new customers than to retain existing ones. Consequently, churn prediction is one of the critical requirements of customer relationship management and customer retention. There had been a number of attempts to predict customer churn, especially in telecom industry. The variables used in churn prediction are both nominal as well as metric in nature. It is well known that certain prediction techniques work well with nominal or ordinal variables where as others work well with metric variables. A hybrid model using classification trees and discriminate analysis is used in this paper to improve the predictions of customer churn in telecom industry.