Boosting Algorithms for Customer Reload Prediction: Optimizing Outcomes with XGBoost, AdaBoost, and CatBoost

The telecommunications industry generates large volumes of data daily, creating significant opportunities for deeper analysis. One of the main challenges faced by companies in this sector is predicting customer reload behavior, where customers decide to top up or reload their services, which directly impacts a company’s profitability. Accurate reload prediction is therefore crucial in helping companies develop effective retention strategies. This study evaluates three machine learning algorithms based on boosting: XGBoost, AdaBoost, and CatBoost, to build a customer reload prediction model. Boosting algorithms are known for their ability to iteratively correct prediction errors, improve model accuracy, and handle complex and imbalanced data. Through experiments conducted on customer data from the telecommunications industry, the results showed that XGBoost achieved an accuracy of 90.04%, AdaBoost 87.02%, and CatBoost outperformed with an accuracy of 88.09%. These findings demonstrate that XGBoost offers the most effective solution for predicting reload behavior, while XGBoost and AdaBoost also provide solid results. This research provides valuable insights for telecommunications companies in identifying customers likely to reload, as well as designing better, data-driven retention strategies.