A Comparative Analysis of Support Vector Machines & Logistic Regression for Propensity Based Response Modeling

Abstract

Increasing cost of soliciting customers along with amplified efforts to improve the bottom-line amidst intense competition is driving the firms to rely on more cutting edge analytic methods by leveraging the knowledge of customer-base that is allowing the firms to engage better with customers by offering right product/service to right customer. Increased interest of the firms to engage better with their customers has evidently resulted into seeking answers to the key question: Why are customers likely to respond? in contrast to just seek answers for question: Who are likely to respond? This has resulted in developing propensity based response models that have become a center stage of marketing across customer life cycle. Propensity based response models are used to predict the probability of a customer or prospect responding to some offer or solicitation and also explain the drivers– why the customers are likely to respond. The output from these models will be used to segment markets, to design strategies, and to measure marketing performance. In our present paper we will use support vector machines and Logistic Regression to build propensity based response models and evaluate their performance.

Keywords: Response Modeling, Propensity, Logistic Regression, Support Vector Machines

View PDF