An optimized stacking ensemble technique for creating prediction model of customer retention pattern in the banking sector

Abstract

Banking is one of the sectors that pays close attention to their clients’ behavior with a view to tracking their activities, most especially as relates to monetary transactions. To add new customers to the existing fold is not only time consuming, but also expensive. This is why Banks generally would like to do everything within their means to ensure the customer retention pattern is consistently high. The objective of this study, therefore, is to create a prediction model that is capable of predicting the retention rate of bank customers. In other to achieve this central goal, this study proposed a machine learning predictive model, created using a function that combines a number of base classifiers to produce an efficient model. The model was created from the dataset retrieved from an open repository, kaggle. The data basically comprised of some demographic and psychological features and the algorithms implemented on these datasets includes: KNN, CART and Naïve Bayes as base classifiers, while the Logistic Regression was used as the Meta Classifier. The model created was evaluated severally to determine its level of accuracy. The resulting output shows a very high accuracy of 83%. A further comparison of this result with the existing related studies unveils that, the proposed ensemble classifier out-performs the existing model which attains 79% to 81% classification accuracies. The proposed model is reliable and can therefore, be used as a bench-mark for similar models created for the prediction of customer retention pattern within the banking sector.