Customer Churn Prediction in Telecommunication Using Gradient Boosting Machine

Machine Learning is being used extensively to solve problems in various fields. Predictive analytics is one of the major applications of machine learning. Predictive analytics uses many techniques, i.e., machine learning, data mining, artificial intelligence, and statistics. It can use present data to make future predictions. Churn prediction is one of the areas where predictive analytics can be applied. The word ‘churn’ can be used in many contexts with different meanings, but in this paper, it refers to a situation when a customer stops using any company’s products or services. Customer Churn Prediction (CCP) is important because the profitability of any company is directly affected by it. It is much costlier to find a new customer than to keep an existing customer who is loyal to the company. If churning customers can be predicted in advance, the company can approach and retain them using some retention mechanism. This paper proposes a churn prediction model that uses the Gradient Boosting Machine (GBM) classifier to identify the customers who may churn. After the classification, the performance of the customer is evaluated using various performance metrics. Accuracy is not a significant metric in performance evaluation as the churn dataset is imbalanced, leading to a misleading result. To overcome this problem, we have considered two more metrics: PR plot and ROC curve.