Analysis of Customer Churn Classification Using Voting Classifier with SMOTE Technique on Telecommunication Data

Abstract

Customer churn is a critical issue in the telecommunications industry because it directly impacts revenue and customer retention strategies. This study aims to evaluate the performance of the Voting Classifier method in churn classification by applying the SMOTE technique to handle data imbalance. The dataset used consists of 7,043 customer data with 19 variables obtained from Kaggle. The research stages include data preprocessing, applying SMOTE to training data, model training using Logistic Regression, Naïve Bayes, and Random Forest with a soft voting approach, and evaluation using accuracy, precision, recall, and F1-score. Evaluations were conducted on several data sharing scenarios to test model stability. The results show that the application of SMOTE consistently increases recall values ​​compared to without SMOTE, especially in scenarios with a larger proportion of training data. The Voting Classifier model with SMOTE achieved the best performance in a 90:10 split with an accuracy of 84.3%, a precision of 80.5%, a recall of 90.5%, and an F1-score of 85.2%. These findings suggest that a combination of ensemble learning and data balancing techniques can improve model sensitivity to minority classes. This research is comparative experimental and can serve as a basis for further development through additional validation and hyperparameter optimization.