Um algoritmo cultural para descoberta de conhecimento em banco de dados

• Ano de defesa: 2011
• Autor(a) principal: Barros, Everton Fernando
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade de Maringá; Brasil; Departamento de Informática; Programa de Pós-Graduação em Ciência da Computação; UEM; Maringá, PR; Centro de Tecnologia
• Palavras-chave em Português: Mineração de dados; Programação genética; Algoritmo cultural; Plano de saúde; Knowledge Discovery in Databases (KDD); Impressões gerais; Algoritmo evolucionário; Brasil.; Data mining; Genetic program; Cultural algorithms; Health insurance; Brazil.; Ciências Exatas e da Terra; Ciência da Computação
• Link de acesso: http://repositorio.uem.br:8080/jspui/handle/1/2581

Health plans, both public and private, have been accumulating large amounts of data containing hidden information that could help their managers to reduce costs of health insurance carriers and assist in planning programs for disease prevention. This knowledge can be discovered using Data Mining (DM) techniques for extracting information relevant and useful. In this context we intend to solve the task of classification that allows the discovery of prediction rules. This dissertation presents an approach to solving the classification task in MD which proposes a hybrid algorithm that uses Genetic Programming (GP) with Cultural Algorithm (CA), which is an evolutionary algorithm based on the process of cultural evolution of humanity. The approach was implemented as a case study data from an operator of health insurance supplement, containing administrative information and procedures in hospitals, laboratories and offices, for the beneficiaries of the state of Santa Catarina. These data were preprocessed and prepared for use by the algorithm. Experiments were performed to evaluate the approach with GP and CA, where CA stores knowledge and helps guide the evolutionary process. A knowledge of these stores general impressions, which are the beliefs of the user, allowing to measure the interest in the rules found. Based on these experiments we evaluated the ability of the algorithm to find rules in two situations: considering the general impressions of the user and without considering the general impression (the user's beliefs) and obtained good results by the two approaches. We also evaluated the ability of the algorithm to find rules from a data set containing small disjoint in which it was possible to find rules which belong to small disjoint and big disjoint, and assess how the general impression of the user can be used for CA.