Enhancing customer churn prediction in telecom services through segmentation

Abstract

Retaining existing subscribers is critical for maintaining profitability in the telecom sector, where acquiring new customers is considerably more expensive than retaining current ones. However, achieving accurate churn prediction remains challenging due to factors such as class imbalance, heterogeneity in customer behavior, and high-dimensional data. Existing studies indicate that traditional churn prediction models often fail to generalize across diverse customer groups, resulting in suboptimal performance. This work proposed an enhanced churn prediction approach that integrates customer segmentation to improve overall predictive accuracy and model efficacy. The K-Means clustering algorithm was proposed to group telecom consumers along with Silhouette Score and Elbow method which are used to determine the optimal number of clusters. For each cluster segment, random forest (RF) and XGBoost (XGB) models were implemented and their performance is compared with traditional non-segmented approaches. To address the data imbalance problem, Synthetic Minority Oversampling Technique (SMOTE) based approach was proposed. While for hyperparameter optimization, the GridSearchCV with 10-fold cross-validation was used. The proposed segmentation approach demonstrated substantial performance improvements for specific customer segments, particularly Clusters 3 and 4, where prediction accuracy exceeded 94%, while maintaining competitive overall performance compared to non-segmented models. These findings highlight the potential of using segmentation-based modeling to enhance churn prediction and support more targeted customer retention strategies in the telecom industry.