Personalized Key Drivers for Individual Responses in Regression Modeling

Abstract

Identification of personalized key drivers is useful to managers in finding a special set of tools for each customer for a better contingency to a higher satisfaction and loyalty and for diminishing risk and uncertainty of decision making. Finding the most attractive attributes of a product for a buyer, or the main helpful features of a medicine for a patient, can be considered via identifying the key drivers in regression modeling. The problem of predictor importance is usually considered on the aggregate level for a set of all respondents. This article shows how to identify a specific set of key drivers for each individual respondent. Two techniques are proposed: the orthonormal matrices used for the relative importance by Gibson and R. Johnson, and the cooperative game theory by Shapley value of predictors in regression. Numerical estimations show that a specific set of key drivers can be found for each respondent or customer, that can be valuable for managerial decisions in marketing research and other areas of practical statistical modeling.