Research on telecom industry customer churn prediction based on explainable machine learning models

Abstract

In the telecom industry, accurate prediction of customer churn is crucial for the companies involved to maintain market competitiveness and increase revenue. To this end, a customer churn prediction framework combining CatBoost algorithm and SHAP model was proposed, aiming to improve the accuracy of prediction and enhance the interpretability of the model. Using the actual business data of a communication company in Xinjiang, the prediction model was constructed through data preprocessing and feature engineering, and five major key performance indicators were selected to evaluate the model performance. The experimental results show that the proposed model outperforms the current mainstream machine learning prediction models in all the above evaluation indicators. Finally, the SHAP framework was introduced to enhance the model interpretability, reveal the key factors affecting customer churn, and provide the specific influence degree of the factors, which provided a scientific basis for telecommunications enterprises to formulate targeted customer retention strategies.

Keywords

• machine learning
• CatBoost algorithm
• SHAP
• predictive model
• telecom industry