Deep Ensemble Classifiers and Peer Effects Analysis for Churn Forecasting in Retail Banking
Modern customer analytics offers retailers a variety of unprecedented opportunities to enhance customer intelligence solutions by tracking individual clients and their peers and studying clientele behavioral patterns. While telecommunication providers have been actively utilizing peer network data to improve their customer analytics for a number of years, there yet exists a very limited knowledge on the peer effects in retail banking. We introduce modern deep learning concepts to quantify the impact of social network variables on bank customer attrition. Furthermore, we propose a novel deep ensemble classifier that systematically integrates predictive capabilities of individual classifiers in a meta-level model, by efficiently stacking multiple predictions using convolutional neural networks. We evaluate our methodology in application to customer retention in a retail financial institution in Canada.