Geração incremental de protótipos controlada por entropia para algoritmos de modelagem preditiva

• Ano de defesa: 2020
• Autor(a) principal: Vasconcelos, Bruno Paulo de
• Orientador(a): Silva, Leandro Augusto da
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Presbiteriana Mackenzie
• Palavras-chave em Português: protótipos; GNG; algoritmos; entropia; acurácia; seleção; redução
• Área do conhecimento CNPq: CNPQ::ENGENHARIAS
• Link de acesso: https://dspace.mackenzie.br/handle/10899/28614

The main proposals of this dissertation are modifying the GNG (Growing Neural Gas) algorithm for prototype generation from a new automatic stop method to find the right amount of prototypes and also the creation of a prototype selection method called KPS with the goal of improving the accuracy in relation to just use the modified GNG. To create this methods were researched the algorithm operation and which techniques are used inside of it. Algorithms like kNN (k Nearest Neighbor), ENN (Edited Nearest Neighbor), DROP3 (Decremental Reduction Optimization Procedure 3), ATISA1 (Adaptive Threshold-based Instance Selection Algorithm 1) and RIS (Ranking-based Instance Selection) were studied in order to make a comparative study with the created methods. The project methodology consists in an exploratory study of the modified GNG and the prototype selection technique with real databases. The full results will be presented in experimental results and soon after will be made the conclusion, noting that the proposed method contributed to the improvement of accuracy.