Prediksi Keputusan Reload Pelanggan Perusahaan Telekomunikasi X Menggunakan Machine Learning

Abstract

The accelerated progress of the telecommunications sector requires
companies to understand consumer behavior in order to maintain
loyalty and increase revenue. The main challenge faced is
managing data to predict customer reload decisions at
telecommunications companies. This study aims to predict these
reload decisions using an Ensemble Learning approach. The
research stages include historical transaction data collection,
preprocessing, handling class imbalance using the SMOTE
technique, feature selection, and model evaluation. The algorithms
used include Random Forest, Gradient Boosting, XGBoost, and
Stacking Ensemble. Model performance was evaluated based on
Accuracy, Precision, Recall, and F1-Score metrics. The results of
the feature selection analysis show that active subscription period
(tenure_rgu) is the most dominant factor influencing customer
reload decisions, followed by the current balance and the history of
total reloads in the last 30 days. In the model evaluation stage,
XGBoost produced the best classification performance with the
highest accuracy rate of 74%, while Gradient Boosting achieved
the highest recall at 50%. The findings of this study are expected to
assist company management in designing more accurate and
efficient marketing strategies to improve customer satisfaction.