Handling imbalanced data in churn prediction using ADASYN and backpropagation algorithm

Abstract

Customers are important assets for an industry, including telecommunication industry. Customer churn will result in lost revenue. For that reason, a classification model that can predict the customer churn is needed. One of the obstacles to develop the classification model is the great difference between number of churn and non-churn customers. The non-churn data dominates the entire customer data. This will make the churn data unfamiliar to the system. ADASYN (Adaptive Synthetic Sampling) is one of the oversampling methods that can be used to solve the class imbalanced problem. ADASYN is algorithm improvement from SMOTE (Synthetic Minority Over-sampling) algorithm [1]. Meanwhile, in this research the backpropagation algorithm is chosen as classification model. The result of the study with proposed scheme shows the testing accuracy 96,31% and testing F1-Score 0.4607.