Uma abordagem multiobjetiva para construção automática de algoritmos de indução de árvores de decisão

• Ano de defesa: 2015
• Autor(a) principal: Silva, Melis Mendes [UNIFESP]
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de São Paulo (UNIFESP)
• Palavras-chave em Português: multi-objective optimizing; evolutionary algorithms; hyper-heuristics; decision trees; otimização multi-objetos; algoritimos evolutivos; híper-heurística; árvores de decisão
• Link de acesso: https://sucupira.capes.gov.br/sucupira/public/consultas/coleta/trabalhoConclusao/viewTrabalhoConclusao.jsf?popup=true&id_trabalho=2108998; https://repositorio.unifesp.br/handle/11600/47320

Decision tree induction is one of the most employed methods to extract knowledge from data, as the representation of knowledge is very intuitive and easily understandable by humans. A successful strategy for inducing decision trees, the greedy top-down approach, has been continuously improved by researchers over the years. After recent breakthroughs in the automatic design of machine learning algorithms, was proposed a hyper-heuristic evolutionary algorithm for automatically generating decision-tree induction algorithms, named HEAD-DT. In this work, this approach was expanded, making the fitness function, which previously worked with only one goal in multiobjective function. In this context, it was used two techniques for optimizing multi-objective fitness: a weighted formula and the lexicographical technique. Experiments will be conducted in 20 public data sets to assess the performance of the new version of HEAD-DT, and we compare it to the traditional decision-tree algorithms C4.5, CART in addition to the original version of HEAD- DT algorithm. Results show that the multi-objective version of HEAD-DT is able to generate promising algorithms when compared to both previous version of HEAD-DT, C4.5 and CART regarding predictive accuracy, F-Measure and complexity (number of nodes). Therefore, this work presents the first efforts to transform the HEAD-DT in a multiobjective algorithm, that is, able to guide the process of evolution by two or more goals.