Customer Churn Prediction in Telecommunication Industry Having Data Certainty

Abstract

Customer Churn Prediction is a challenging activity for decision makers because most of the time, churn and non-churn customers have similar features. It is one of the major concerns for large companies, especially in the field of telecommunication field. Churn can be considered as a binary classification. The classifiers shows different accuracy levels at different zones of data. In such cases, a correlation can easily be observed in the level of classifier's accuracy and certainty of its prediction. So a mechanism to estimate the classifier’s certainty for different zones within the data is needed so that the expected classifier’s accuracy can be estimated. Here the classifier’s certainty estimation is done using six sigma rule of normal distribution applied on the correlation values of all features in the dataset. Based on this the dataset is grouped into two categories such as (i) data having high certainty, and (ii) data having low certainty. Based on these criteria, classifier accuracy is estimated in the high distance zone. From the different evaluation measures like accuracy, f-measure, precision, recall and Receiving Operating Characteristics (ROC) area, the performance of classifier is evaluated. Then by applying a k fold approach the certainty of the classifier decision is estimated.