Comparative Analysis of Two Classifiers
Implementing NOMINAL Logistic Regression

Hélio Radke BITTENCOURT[1]

Bárbara Patrícia Olberman PASINI1

Denis Altieri MORAES[2]

Bruno Duarte dos SANTOS[3]

Vitor HAERTEL[4]

§    ABSTRACT: There is a wide range of statistical classification methods described in the scientific literature. Each method presents advantages and disadvantages that depend mainly on the probabilistic behaviour of the classes. This article presents two methods of classification based on logistic regression and shows results obtained with three real data sets. The first method is a single-stage classifier based on Nominal Logistic Regression. The second method is a hierarchical classifier, structured like a binary decision tree that uses the traditional logistic regression at each stage. Both methods are compared with traditional classifiers and the results are discussed.

§    KEYWORDS: Nominal logistic regression; decision trees; classifier; pattern recognition.

 



[1] Departamento de Estatística, Faculdade de Matemática, PUCRS, Av. Ipiranga, 6681 CEP: 90619-900, Porto Alegre, Brazil, and Programa de Pós-Graduação em Geografia, UFRGS, Brazil. E‑mail: heliorb@pucrs.br / barbarap@pucrs.br

[2] Departamento de Estatística, UFSM, Av. Roraima, 1000, CEP: 97105-900, Santa Maria, RS, Brazil, E‑mail: d_altieri@yahoo.com.br

[3] Bolsista de Iniciação Científica, Pibic– PUCRS, Porto Alegre, RS, Brazil. E‑mail: brunoduantos@hotmail.com

[4] Centro Estadual de Pesquisas em Sensoriamento Remoto e Meteorologia, UFRGS, Porto Alegre, RS, Brazil, Caixa Postal: 15044, CEP: 91501-970, Porto Alegre, RS, Brazil, E‑mail: victor.haertel@ufrgs.br