Analysis of an Optimized LightGBM Based on Different Objective Functions for Customer Churn Prediction in Telecom

Telecommunication industry is one of the highly competitive sectors with high customer churn rates. Customers are important to a business for its sustainability. Therefore, businesses are endeavouring to reduce the churn rate since the cost of acquiring a new customer is much higher than retaining an existing customer. In this paper, a customer churn prediction framework is developed with an optimized LightGBM algorithm, coined eLightGBM. The optimization is based on different factor awarenesses defined in different objective functions by using OPTUNA optimization. In eLightGBM, the model hyperparameter can be fine-tuned based on different objective functions, which are recall score maximization and F1 score maximization. Specifically, LightGBM hyperparameters are adjusted in each repeated loop until a satisfactory recall or F1 is scored. The empirical results demonstrate that the proposed eLightGBMs show higher recall scores compared to the other machine learning classifiers. This indicates that eLightGBMs are able to detect a larger proportion of churners.