Telecom customer churn prediction in context of composite model

For the increasingly saturated telecommunications market, the way to determine whether customers will churn and find the reasons that affect customer churn are the two main factors affecting the operation of telecom companies. In this study, a combinatorial model prediction composed of logistic regression and neural networks was adopted. It contains both the efficient and precise characteristics of machine learning and the good explanatory nature of logistic regression. The prediction accuracy of the final model reached about 80.5%, which is about 3% lower than that of a single neural network model and about 7% higher than that of logistic Regression. According to the results of the significance test, the main factors affecting customer churn are the following aspects: long-term costs, service duration, network services, and contract types. The same approach of this study can be applied to other industries (e.g., video site membership), allowing decision makers to predict the flow of customers and develop plans in advance to reduce losses. These results shed light on guiding further exploration of an analytical approach to contemporary customer churn and methods of decision-making.