Projection of Customer Churn in Telecom Sector through Machine Learning Algorithms on Big Data Platforms

The telecom industry study is essential for increasing the profitability of enterprises, especially through precise churn prediction. The goal of this study was to create a specialized churn prediction system for the telecom provider SyriaTel. For accurate churn estimates, high AUC values were necessary, and the dataset was divided into 30% testing and 70% training sets. Hyperparameter adjustment and accurate model evaluation were made possible via cross-validation. To get the features ready for machine learning algorithms, feature engineering and selection techniques were used. Tree-based methods and under-sampling were used to address data imbalance. Decision Tree, Random Forest, Gradient Boosting Machine, and XGBOOST are the four tree-based models that were selected. Strategic planning and the incorporation of mobile social network features were essential to success. With a 93.301% AUC on the SyriaTel dataset, XGBOOST performed better than GBM, Random Forest, and Decision Tree. When tested on a fresh dataset, XGBOOST's AUC was 89%. Non-stationary data necessitates frequent model retraining. Social network analysis was used to improve telecom churn prediction.