Júlio Michael Stern

Universidade de São Paulo
Instituto de Matemática e Estatística
Departamento de Ciência da Computação
Caixa Postal 66281
CEP 05315-970 São Paulo, SP 

Abstract

This work presents REAL, a Real-Valued Attribute Classification Tree Learning Algorithm. Users demands for a decision support tool explain several of the algorithm's unique features. Compared to competing algorithms, in our applications, REAL presents major advantages:

1- The REAL classification trees usually have smaller error rates. 
2- A single conviction (or trust) measure at each leaf is more convenient than the traditional (probability, confidence-level) pair. 
3- No need for an external pruning criterion.

Co-authors: Fábio Nakano and Marcelo Lauretto