A0166
Title: Ordinal regression for non-ordinal data? It's all about parsimony!
Authors: Yang Ni - Texas AM University (United States) [presenting]
Abstract: The aim is to introduce a novel regression model for categorical data termed classification with optimal label permutation (COLP). By design, COLP is a more parsimonious classifier than multi-class logistic regression and hence can have better out-of-sample prediction performance. The idea is simple - the class label is ordered so that ordinal regression is applicable. While label order does not have to have any physical meaning, it sometimes does. In addition to classification, interestingly, COLP gives rise to a provably identifiable causal graphical model for categorical data, whereas multi-class logistic regression does not.
