Título: Hybrid Neural Solutions for Automatic Knowledge Discovery from Databases
Autores: Amorim, Bruno P.; Vasconcelos, Germano C.; Brasil, Lourdes M.
Resumo: Artificial Neural Networks (ANN) have been successfully used in a wide variety of real-world applications. However, ANN alone have not been fully employed in KDD (Knowledge Discovery in Databases) applications because they often produce incomprehensible models. Neuro-fuzzy systems and techniques for symbolic knowledge extraction have been increasingly used to represent the knowledge acquired by ANNs in a comprehensible form. This paper presents hybrid neural solutions for the KDD process, resulting from a detailed experimental investigation of three neural models (MLP, FuNN and FWD), four symbolic knowledge extraction techniques (AREFuNN, REFuNN, TREPAN and FWD) and two feature selection algorithms (FWD and the decision tree extracted by TREPAN). A large scale credit assessment application in a real-world situation was used as the test bed for the experimental investigations carried out. The results demonstrate that the benefits obtained from hybrid neural solutions are actual.
Código DOI: 10.21528/CBRN2005-099
Artigo em PDF: CBRN2005_099.pdf
Arquivo BibTex: CBRN2005_099.bib