Understanding telecommunication customer churn : insights from LightGBM predictive modelling and SHAP feature interpretation

Manuscript Type: Research Article Research Aims: This research is expected to contribute to scientific insights on machine learning predictions, how to understand the contribution of variables and their interpretation in the overall context of customer churn in telecommunications companies. Design/Methodology/Approach: Data was collected from one Indonesian telecommunication company within a total of 50,000 sample data points. The data was analyzed using a machine learning algorithm to process, predict, and interpret the result based on the research scenario. Research Findings: These findings revealed that the SHAP framework significantly impacts the churn problem, allowing the marketing team to implement the right strategy based on customer personalization. Theoretical Contribution/Originality: This research enriches customer churn research in the telecommunications industry by introducing a combined method of the LightGBM model and SHAP framework, providing a thorough analysis of variable contributions to the customer churn model used, and addressing the lack of research that focuses on variable contributions. Practitioner/Policy Implication: This research provides an overview of the utilization of customer variables that can later be studied deeply by data or marketing teams to produce initiative projects based on data and machine learning models. Research Limitation/Implication: Future studies could combine the feature selection method to filter the model's features and remove redundant ones, thereby producing an analysis of the contribution of variables that truly impact customer churn.