Selection of multinomial logit models via association rules analysis

In this research, we propose a novel approach for a multinomial logit model selection procedure: specifically, we apply association rules analysis to identifying potential interactions for multinomial logit modeling. Interaction effects are very common in reality, but conventional multinomial logit model selection methods typically ignore them. This is especially true for higher-order interactions. Here, we develop a model selection framework to address this problem. Specifically, we focus on building an optimal multinomial logit model by (1) exploring the combinations of input variables that have a significant impact on response (via association rules analysis); (2) selecting potential (low-order and high-order) interactions; (3) converting these potential interactions into new dummy variables; and (4) performing variable selections among all the input variables and the newly created dummy variables (interactions). Our model selection procedure is the first approach to provide a global search for potential interactions and establish the optimal combination of main effects and interaction effects in the multinomial logit model. In our investigation, we consider both simulated and real-life datasets, thereby confirming the effectiveness and efficiency of this method.