Machine learning approach to predict purchase decision of bank products and services

We propose a machine learning approach to predict purchase decision of bank products and services. The data were collected from May 2008 to May 2014 of a Portuguese bank. This investigation will help to predict the business of the bank and financial inflation and recent trends in bank product and services. The investigation is focused on the classification and prediction of bank telemarketing calls for term deposit product. We have analysed a large dataset of 41,188 observations related with bank client, product, services and socioeconomic attributes. Initially, the dataset was having 150 features and we have selected 21 most relevant features using standard adaptive forward selection and intelligence quotient. We have also compared four machine learning approaches: conditional inference trees (Ctree), recursive partitioning (Rpart), support vector machines (SVM) and random forest. The paper contains an impact analysis of changing training dataset and training time of a model. Observatory study shows the integration of both parameters: accuracy and model learning time to form a generalised and optimised solution for predicting bank business.