Multinomial logit model building via treenet and association rules analysis: An application via a thyroid dataset

Abstract

A model-building framework is proposed that combines two data mining techniques, TreeNet and association rules analysis (ASA) with multinomial logit model building. TreeNet provides plots that play a key role in transforming quantitative variables into better forms for the model fit, whereas ASA is important in finding interactions (low-and high-order) among variables. With the implementation of TreeNet and ASA, new variables and interactions are generated, which serve as candidate predictors in building an optimal multinomial logit model. A real-life example in the context of health care is used to illustrate the major role of these newly generated variables and interactions in advancing multinomial logit modeling to a new level of performance. This method has an explanatory and predictive ability that cannot be achieved using existing methods.