Bw-Xgboost-Nn: A Novel Customer Churn Prediction Framework For The Retention Of Churning Customers

Abstract

Since the cost of attaining a novel customer is more expensive when compared with retaining churning customers, customer retention is a vital process for a company. But, as the data collection grounded on reasons for churning along with processing is a challenging task, Customer Churn Prediction (CCP) is a complex process. Although more research has been conducted on the Churn Prediction (CP) process, enhancement is still required for accurate prediction. Thus, to enhance the CCP model’s output, this paper proposes a novel Binary Whale-eXtreme Gradient Boosting-Neural Network (BW-XGBoost-NN) technique. The headers are read from the dataset initially; then, utilizing the Hadoop Distributed File System (HDFS), the repeated data are removed. After that, the data is pre-processed and the Modified Barnacles Mating Optimization Algorithm (MBMOA) selected the features as of the pre-processed output. Lastly, utilizing the BW-XGBoost-NN classifier, the churn data are predicted as churn and non-churn. The churning customer can be retained grounded on the predicted output. For proving its efficiency, the proposed framework is experimentally analyzed regarding the accuracy, Feature Selection (FS) time, fitness vs iteration, recall, et cetera. The outcomes exhibited that the proposed mechanism outperforms the prevailing models; also, the predicted output level is given.