Analysis Implementation of the Ensemble Algorithm in Predicting Customer Churn in Telco Data: A Comparative Study

Abstract

Globalization and technological advancements in the telecommunication industry have led to a significant rise in the number of operators, leading to intense market competition. This sector has become crucial in developed countries, and companies strive to increase profits by acquiring new customers, up-selling existing ones, and extending the retention period of current clients. In the traditional method of defect prediction, a single classifier is used to build a model on a pre-labeled dataset. However, this approach has limitations in predicting defects accurately under certain circumstances. To overcome these limitations, boosting is applied to combine multiple weak classifiers and create a robust classification model. Among many algorithms used for churn prediction, ensemble techniques have demonstrated greater accuracy than simpler approaches. This study aims to overcome these limitations by experimenting with five ensemble algorithms, including Adaboost, Gradient Boost, XGBoost, CatBoost, and LightGBM. The results indicate that XGBoost outperforms other techniques and is the most suitable algorithm to build the predictive model. Additionally, the study achieves higher accuracy by performing a Grid Search CV hyper-parameter setting with XGBoost, resulting in an accuracy of 81.2%.