Machine Learning Based Customer Churn Prediction in Banking Sector

Customers in 21st century have access to a wide range of ways to deposit money, both online and offline, which leads to constant customer churn for the whole banking industry. In order to retain existing customers, the bank sector has been prioritizing building models which aim to predict clients who may exit in the future. In this paper, based on machine learning techniques, different models such as XGboost, Catboost and LightGBM are fitted to the churn modelling dataset from Kaggle, contributing to the prediction of potential bank customer churn. In addition, some methods of feature selection and hyperparameter tunning are used to enhance the performance of final prediction results. The results generated by different models are compared in terms of accuracy, precision, recall, etc. Age and the number of products purchased from the bank are suggested to be 2 factors that greatly influence the prediction results. LightGBM model shows the best general performance and therefore is recommended for future prediction.