Churn Prediction in Telecoms Using a Random Forest Algorithm

The telecoms industry has been exposed to increased churn rates compared to other industries. Over the years, there has been a significant drive to construct systems that will identify customers that exhibit the potential to churn. Churn prediction systems utilize various algorithms that are applied to customer datasets to determine customers that will potentially churn. This paper is based on the development and application of a churn prediction system using aRandom Forest algorithm. The system was applied to a publicly available dataset that has been used in related studies as well a real customer dataset from a telecoms organization in South Africa. The model highlighted that data sampling and hyperparameter optimization are critical steps in the development of churn prediction systems. The model when applied to both datasets achieved greater than ninety percent (90%) accuracy. Therefore, this model has demonstrated that future prediction models should be applied and measured utilizing a balanced and hyperparameter approach.