Project With Haolun Shi

Tree-Based Methods for the Multinomial Logit Model

We consider developing tree-based methods for the multinomial logit model to estimate the utilities/choice probabilities nonparametrically. Ideally, our aim is to develop a method that can be used for discrete choice modeling and accurately predict the choice probabilities. Our tree-based procedure should possess properties similar to the random forest, and is able to avoid over-fitting via carefully tuning the parameters. We will conduct numerical studies to evaluate the usefulness of our method, in terms of prediction and recovering the covariate importance.