| |
| Acadêmico(a): Alexandre Daniel Dalabrida |
| Título: Protótipo de um Sistema de Extração de Regras Simbólicas de Redes Neurais Artificiais Utilizadas na Tarefa de Classificação em Data Mining |
| |
| Abstract: |
| This work has a main goal to evaluate the applicability of rule extraction techniques over the artificial neural nets, more specifically in the data mining task of data classification. To evaluate this applicability was implemented an archetype of data mining system specifically for the classification task based on artificial neural net, training the same to classify data for external data sources. Through applied specific rule extraction algorithms over the trained net, had been extracted and articulated more appropriate rules of classification to the human agreement. In the construction of the archetype, had been analyzed techniques and tasks of data mining, artificial neural nets and on extraction of rules of neural nets. As consequence of the development of this work, was demonstrated that the extraction of the implicit rules in the neural net assists clearly in the agreement of the model of classification of data, evidencing the hypothesis inferred for the artificial neural network of a clearer form to the agreement. |
| |
|
