Book Description

Owing to fierce competition among telecom companies, customer churn is inevitable and it is known that the cost of retaining existing customers is 5-10 times lower than the cost of obtaining new customers. The competitive telecom industry requires telecom companies to use Customer Relationship Management (CRM) to analyze customer churn. CRM analysts need to predict customer churn and understand the reasons for churn.

In this dissertation, a CRM system which consists of an ensemble prediction system and a customer churn analysis system is proposed.   An ensemble prediction system is composed of a stacking model and soft voting. XGBoost, Logistic Regression, Decision Tree, and Bayesian machine learning algorithms are used to develop a stacking model with two levels, and the three outputs of the second level are used for soft voting. The feature construction of the churn dataset is a process which builds new features from the original dataset by the grouping of customer behavior features to increase the feature space and discover potential information. Using four evaluation measures, the stacking model is examined with the original and new churn datasets. The results show that the proposed customer churn analysis system achieves 96. 1% and 98% accuracy for the original churn dataset and the new churn dataset, respectively. And it has a much better performance compared to other recent search results.

The feature importance based churn analysis system to reduce customer churn and provide customized strategies and services for a telecom company is designed, which is composed of three parts: clustering analysis, feature importance ranking and data visualization. RFM model and K -means algorithm with an elbow method are used to make clusters from the original dataset and the XGBoost algorithm with SHAP method is used to get a feature importance ranking. And the simulation results show that the K value of the K -means algorithm is selected to 4 and dataset is divided into four groups. And each cluster has a different feature importance ranking so that the specialized strategies can be provided to each cluster.

The proposed customer churn analysis system will be able to enhance the ability of telecommunication companies to develop and implement effective CRM and marketing strategies.