Model Selection for Multinomial Responses via Association Rules Analysis

• Activity Number: 467
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Computing
• Author(s): Pannapa Changpetch*+ and Dennis Lin
• Companies: Penn State University and Penn State University
• Keywords: Multinomial logit model ; Association rules analysis ; Model Selection

Abstract

In this paper, we propose a model selection procedure for multinomial responses by implementing association rules analysis. In classical way, multinomial logit model building usually consider only main effects and ignore interactions. Here, we develop a model selection framework that is particularly important for applications in which interactions are present. We do this by (1) find (low-order and high-order) interactions among input variables (via association rules analysis) (2) select potential interactions to be considered in model building (3) convert these potential interactions into new dummy variables, and (4) perform variable selection among all input variables and the newly created dummy variables (interactions) to build up the optimal multinomial logit model.