Abstract
Numerous studies and research work has been undertaken in the area of creating predictive models for studying Bank Churn. In these studies, the end goal was to create a high accuracy predictive model; while this is commendable, this research focuses on creating an architecture for a predictive model by aggregating the power of various predictive models. The architecture and model proposed in this paper achieved an accuracy of 91% in the test data (35% of the original data set), and an AUC of 96% - confirming the generalized nature of the model. Also, various feature extrapolation techniques were introduced which provide valuable insights to the banking sector.