Efficient feature selection framework for digital marketing applications

Abstract

Digital marketing strategies can help businesses achieve better Return on Investment (ROI). Big data and predictive modelling are key to identifying these specific customers. Yet the very rich and mostly irrelevant attributes(features) will adversely affect the predictive modelling performance, both computationally and qualitatively. So selecting relevant features is a crucial task for marketing applications. The feature selection process is very time consuming due to the large amount of data and high dimensionality of features. In this paper, we propose to reduce the computation time through regularizing the feature search process using expert knowledge. We also combine the regularized search with a generative filtering step, so we can address potential problems with the regularized search and further speed up the process. In addition, a progressive sampling and coarse to fine selection framework is built to further lower the space and time requirements.