Predicting Customer Churn in Telecommunications with Machine Learning Models

Customer churn is an important issue in businesses and service sectors including telecommunication industries. Prediction of potential customer churn can be very useful in these fields and it can help to improve customer retention significantly by providing personalized offering to reduce potential churn. This paper mainly focused on customer churn prediction using machine learning (ML) models and Iranian Telco customer churn dataset. Different reasons or variables are involved in customer dissatisfaction or indication of customer’s churn. If ML models are trained with such essential and crucial data, they can provide better prediction with high accuracy. This study experimented 16 ML models (single as well as ensemble) and the model performances were evaluated in two ways - 80:20 train-test split and 10-fold cross-validation. In 80:20 split, categorical boosting CatBoost classifier outperformed other models with 97.54% accuracy. However, Light Gradient Boosting Machine classifier LGBMC performed best in 10-fold cross-validation by achieving average accuracy of 96.39% with 1.08% standard deviation while CatBoost classifier was second best performing model. Thus, machine learning techniques can very effectively serve Telco industry as well as other customer-oriented competitive businesses. However, limitation of this study is that it experimented one dataset having 2850 distinct instances with 14 feature variables including target feature. Hence, further experiments with more and large datasets are suggested. Businesses including telecommunication industries can implement these ML models by deploying them in cloud or online platform. In this regard, an online database is required to maintain to automatically collect the from customers involving all the feature variables. If potential churn of any customer is predicted by the ML models, initiatives should be taken by providing improved customer care and personalized services to retain the customer.