Advanced Machine Learning Methods for Churn Prediction and Classification in Telecom Sector

Abstract

The telecom industry’s vast client base causes it to create massive amounts of data daily. The cost of obtaining new clients is higher than the cost of keeping existing ones, according to business experts and decision-makers. Customer churn remains a critical challenge in the telecom sector, where retaining existing customers is significantly more cost-effective than acquiring new ones. This study focuses on applying advanced machine learning techniques to accurately predict and classify customer churn using the Telco Customer Churn dataset. The methodology includes comprehensive data preprocessing such as handling missing values, converting data types, encoding categorical features, scaling, and addressing class imbalance using SMOTE. The dataset is split into training and testing sets, and models including XGBoost, Extra Trees, and a Voting Classifier are developed and fine-tuned using Randomized SearchCV. Evaluation is conducted using accuracy, precision, recall, F1-score, ROC-AUC, confusion matrices, and classification reports. Among the models tested, the Extra Trees classifier achieved the highest performance with $86.05 \%$ accuracy, recall, and F1-score, outperforming both the Voting Classifier and XGBoost. Additionally, it surpassed baseline models like KNN, Logistic Regression, SGD, and Random Forest, and demonstrated more balanced performance compared to the stacking model, which, despite delivering competitive results, exhibited lower recall and less consistency across evaluation metrics. The results confirm that the Extra Trees model provides a robust and balanced solution for churn prediction, offering valuable insights for telecom providers in developing proactive customer retention strategies.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 7 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.