Churn Prediction in the Banking Sector

Abstract

Customer churn prediction is extremely important for any business as it recognizes the clients who are likely to stop using their services. In our proposed system, a bank dataset is used which predicts possible customers that are likely to churn. The dataset consists of 10,000 rows and 13 features. Retaining existing customers is much cheaper than finding new customers. Hence, many firms have set reducing churn as their main priority for which they try to understand why customers churn and eliminate that risk. Churn prevention is useful in all industries which have customer subscription, as their cancellations are clearly observed. This also helps non-contractual businesses where they face a challenge of defining a clear churn event timestamp. It is difficult to build a model which will predict customer churn in the banking sector as customers don’t sign a contract regarding the duration of services with the bank. Building a strong customer relationship with their customers is the prime concern for many organizations of various sectors like the banking sector, airline ticketing services, financial services, telecommunications, and ott platforms like Netflix. This will bring profits to companies. Therefore, predicting if a customer will churn is an extremely effective task for many companies. In the proposed system, a model is developed using the ANN (Artificial Neural Network) technique like backpropagation and optimization technique stochastic gradient descent algorithm to predict the customers who are likely to churn using past information and behavior. Through feature extraction, features which influence customers to churn were found. The accuracy which we have obtained through our model is 86.4%. We have obtained a precision value as 95.48% and the recall value as 88.39%.