Customer Churn Prediction Using Machine Learning Technique With Certainty Computation
Rapid growth of the digital data, industries become hastily digitalized. The Telecom industry is one of the most revenue-generating ones. Here the telecom industry faces customer churn problems. Customers who were wanted to switch their network to another network. Predicting these customers and providing a better option to customers will help balance the revenue of their industry. Customer Churn Prediction (CCP) is a confronting activity for machine learning and decision maker's community. Non-churning and churning customers hold some similar kind of resembling features. With various experimentation over customer churn and corresponding data, it is identified that classifier provides various accuracy levels for various dataset zones. In some situations, the correlation among features is observed easily in various classifier accuracy and prediction certainty. In this case, if any mechanisms are defined to evaluate classifiers certainty in various zones along with data, then predictable classifier accuracy could be evaluated before classification. Here, a novel prediction model is provided based on validating classifier certainty using a distance measure. Datasets are clustered into various zones based on distance measures that are classified as two categories: data with higher certainty level and data with lower certainty level for predicting behavioral nature of churn and non-churn customers. A distance measure will perform with the gaming theory. Performance measures like accuracy, precision, recall, and f-measure on diverse publicly available telecom industry datasets are computed to demonstrate the classifier's certainty level. Classifier acquires higher accuracy with distance measure, that is, churn/non-churn with higher and lower certainty.