Assessing the Predictive Performance of Machine Learning in Direct Marketing Response

Abstract

This paper intends to better understand the pre-exercise of modeling for direct marketing response prediction and assess the predictive performance of machine learning. For this, the authors are using a machine learning technique in a dataset of direct marketing, which is available at IBM Watson Analytics in the IBM community. In the results, first, among all variables, customer lifetime value, coverage, employment status, income, marital status, monthly premium auto, months since last claim, months since policy inception, renew offer type, and the total claim amount is shown to influence direct marketing response. However, others have no significance. Second, for the full model, the accuracy rate is 0.864, which implies that the error rate is 0.136. Among the patients who predicted not having a direct marketing response, the accuracy that would not have a direct marketing response was 87.23%, and the accuracy that had a direct marketing response was 66.34% among the patients predicted to have a direct marketing response.