Scholarlink

Customer churn remains a critical challenge in the telecom industry, where retaining existing subscribers is significantly more cost-effective than acquiring new ones. This study presents a machine learning-based framework for churn prediction using ensemble classifiers such as XGBoost and LightGBM, alongside traditional models like Random Forest and Decision Tree. The dataset underwent extensive preprocessing, feature engineering, and class imbalance handling using Borderline-SMOTE. Recursive Feature Elimination (RFE) has been used to identify the most informative attributes, while GridSearchCV optimized hyperparameters. Among the evaluated models, XGBoost achieved the highest performance with an F1-score of 0.92 and cross-validated accuracy of 99.16%. SHAP analysis provided transparency by identifying key churn predictors. Comparative evaluations using ROC curves and confusion matrices validated the model’s robustness. The proposed solution offers high predictive accuracy and actionable business insights to support customer retention strategies.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Optimizing Telecom Churn Forecasting Using XGBoost and LightGBM with Data Balancing and Feature Refinement

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Shreya Tiwari,
  • Lakshya Gupta,
  • Himanshu Verma

摘要

Customer churn remains a critical challenge in the telecom industry, where retaining existing subscribers is significantly more cost-effective than acquiring new ones. This study presents a machine learning-based framework for churn prediction using ensemble classifiers such as XGBoost and LightGBM, alongside traditional models like Random Forest and Decision Tree. The dataset underwent extensive preprocessing, feature engineering, and class imbalance handling using Borderline-SMOTE. Recursive Feature Elimination (RFE) has been used to identify the most informative attributes, while GridSearchCV optimized hyperparameters. Among the evaluated models, XGBoost achieved the highest performance with an F1-score of 0.92 and cross-validated accuracy of 99.16%. SHAP analysis provided transparency by identifying key churn predictors. Comparative evaluations using ROC curves and confusion matrices validated the model’s robustness. The proposed solution offers high predictive accuracy and actionable business insights to support customer retention strategies.