Product partition model for categorical features

• Autor(a) principal: Tulio Lima Criscuolo
• Data de Publicação: 2019
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Institucional da UFMG
• Texto Completo: http://hdl.handle.net/1843/34069
• topic: Hierarchical Model; Regression; Clustering; Dimensionality Reduction; Modelo Hierárquico; Regressão; Clusterização; Redução de dimensão

A common difficulty in data analysis is how to handle categorical predictors with a large number of levels or categories. There are few proposals developed in the literature to handle this important and frequent problem. We introduce a generative model that simultaneously carries out the model fitting and the aggregation of the categorical levels into larger groups. Our approach is based on imposing a graph where the nodes are categories and creating a probability distribution over meaningful partitions of this graph. Being a Bayesian model, it allows the posterior inference, including uncertainty measurement, on the estimated parameters and the categories partition. We compare our method with the state-of-art methods showing that it has equally good predictive performance and much better interpretation ability. Given the current concern on balancing accuracy versus interpretability, our proposal reaches an excellent result.