A fuzzy based churn prediction and retention model for prepaid customers in telecom industry

Accurate and timely identification of the potential churner, also known as churn prediction, is crucial to devise effective retention strategies. A number of churn prediction models have been proposed in the past, however, the existing models suffer from a number of limitations due to which these models are not applicable on real world large size telecom datasets. Firstly, the feature selection methods adopted in majority of the previous models neglected the information rich variables present in Call Details Records (CDRs). Secondly, the present models have been validated mainly with benchmark datasets which don’t provide a true representation of a large scale real world data in telecom sector. Thirdly, there is very limited amount of work reported in literature that has extended the prediction models towards automatic and intelligent retention mechanisms. Moreover, categorization and severity of the predicted churners has not been focused in the past for targeted and intelligent retention campaigns. Motivated by the aforementioned limitations, we propose a novel churn prediction and retention model for achieving the aim of accurate identification and targeted retention of churners. Primarily, our contribution is the accurate identification of churners at different severity levels using fuzzy based classifiers. Secondarily, our model automatically generates intelligent retention campaigns by mining customer usage and complaints patterns. Experimental results on real world telecom data of a South Asian company revealed the supremacy of fuzzy classifiers in terms of classification by achieving 98% accuracy of churner class. Moreover, the proposed retention strategy based on churner severity and categorization managed to retain 87% of the potential churners.