Big Data and Machine Learning for Forestalling Customer Churn Using Hybrid Software

Abstract

The term customer churn is used to describe a situation where a customer leaves one merchant or supplier and moves to a competitor of that original merchant or supplier. This is also know as customer attrition. Prior to churning, however, there are often hints or clues in the customer’s buying patterns that he or she is ready to leave the supplier. This paper looks at the use of Machine Learning algorithms to predict when customers are ready to churn or in the process of churning. These predictions are then used to look at free text unformatted log data to find any reasons why this customer might be churning. This free text log data would include textual error messages that the customer might have received or financial problems which might have arisen such as not having sufficient funds in his or her account. This merged information is then forwarded to an outbound call queue so that trained call center agents could make human-to-human voice calls to the customer and entice them to stay with the merchant or supplier by offering some financial incentive. All of the technicalities were orchestrated using Spring Boot microservices. Design Science Research was used for the this project and a number of iterations were executed until results were satisfactory. These iterations included changing from an AutoEncoder to a MultiLayerPerceptron, included changing from one Java library providing neural network objects to another Java library, included better searching of log files for possible reasons that customers were churning and included many experiments with the quantity of sales data required in order for the neural networks to create reasonable predictions.