Prediction Model for Customer Churn Using Data Mining Techniques

Abstract

Customer churn or customer attrition is the phenomenon where customers of a newlineSpecific business discontinues buying or communication with the organization. A high newlineChurn means that a higher number of customers no longer want to buy goods and newlineServices from the business. Customer churn rate or customer attrition rate is the newlinemathematical calculation of the percentage of customers who are not likely to make newlineanother purchase from a business. Customer churn happens when customers decide to newlinenot continue purchasing products/services from an organization and end their newlineassociation. It is an integral parameter for the organization since acquiring a new newlinecustomer could cost almost many a times more than retaining an existing customer. newlineThe churn customers are a heavy loss for the enterprise and business industry. A newlinecustomer who is loyal and using a particular product and service for a long time, newlinesuddenly takes a decision to change their mind set to use another product and service. newlineIf the industries or enterprises know that these customers are planning to migrate and newlinewant to use another product and services in that situations they can use business newlinestrategies to protect them. This research work is related to a predictive model on churn newlinecustomer based on their credit score, history of buying patterns, geography location, newlinegender specification and estimated salary with proper balance. The proposed newlinepredictive model is based on the 10K customers data which were collected from the newlinedifferent geographical location. The predictive Models Used: Logistic Regression, newlineDecision trees, Gradient Boosting, XGBoost as base learners. The results of this churn newlinepredictive model is compared with that of the logistic regression, SVM, Gradient newlineBoosting, Random Forest algorithms. The predictive model is based on hybrid newlineXGboost with SMOTE which is used for management of imbalanced data and also newlinehypertuned and is having accuracy of 96%. Here, it is assured that this predictive newlinehybrid model could be utilized in the industries which are service oriented in newlinepre