Integração de técnicas de raciocínio baseado em casos e agrupamento de dados na construção de sistemas inteligentes

Abstract

Case-Based Reasoning (CBR) systems use past experiences, or cases, to make decisions about current problems. To retrieve these case base experiments, similarity techniques are employed on the search for past cases similar to the current query problem. In these systems, problems related to the relevance analysis of attributes that represent information recorded in cases can impact the performance of CBR case retrieval mechanisms. In this context, this dissertation uses clustering algorithms applied on cases to identify which are the most relevant attributes represented in CBR case retrieval, which can be exploited as indexes in the construction of similarity functions adjusted for the solution of application problems. In this process, different clustering algorithms and evaluation metrics of clustering cases are investigated in an indexing framework, allowing to identify which are the most relevant attributes. This dissertation also uses clusters formed in clustering to create subcase's bases to be used by CBR query engines. To evaluate the proposed approach, a study case involving a simulation system is explored. Results obtained in this study case demonstrate that the use of the framework proposed in this dissertation allows to improve the accuracy of CBR queries from 44.50% to 83.93%. Web case bases are also explored in the validation of the proposals presented in this dissertation.