A Novel Approach to Customer Churn Prediction in Telecom

Customer turnover constitute a remarkable challenge for large companies, particularly in the telecom sector, impacting their revenues. To address this, there's a need to improve a model predicting potential customer churn. Identifying the problems that contribute to customers leaving is essential for implementing effective measures. The primary focus of this study initiates with an examination of the Telecom Churn dataset, encompassing customer details such as analytics and patterns. The dataset includes a label indicating whether a customer has churned or not. Through data analysis, we aim to uncover correlations, patterns, and potential predictive elements relevant to churn prediction. Our goal in this study is to create a strong ensemble learning setup, bringing together numerous classification methods whichincludes Logistic Regression, K-Nearest Neighbors, Light Gradient Boosting Machine (LGBM),eXtreme Gradient Boost (XGB), Histogram based GradientBoost (HISTGB), and Categorical Boost (CatBoost).Excitingly, our focus on HistGB showed impressive results, achieving a notable ROC AUC accuracy of 99.49%. This success highlights HistGB's effectiveness in our classification task. Additionally, we use a soft ensemble method, boosting predictive power by smartly combining outputs from individual models.Remarkably, our ensemble approach achieved the highest accuracy rate of 96.39%. This careful blending forms a solid foundation for a more reliable and accurate predictive framework, demonstrating the potential of our approach to enhance classification performance.