Desenvolvimento e aplicação de heurística para calcular pesos e bias iniciais para o "back-propagation" treinar rede neural perceptron multicamadas

• Ano de defesa: 2017
• Autor(a) principal: Silva, Aldemário Alves da
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal Rural do Semi-Árido; Brasil; Centro de Ciências Exatas e Naturais - CCEN; UFERSA; Programa de Pós-Graduação em Ciência da Computação
• Palavras-chave em Português: Aprendizado de máquina; Rede neural perceptron multicamadas; Heurística; Reconhecimento de padrões; Classificação de dados; Machine learning; Neural network multilayer perceptron; Heuristic; Pattern recognition; Data classification; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufersa.edu.br/handle/prefix/5215

The training of Multilayer Perceptron Neural Network (MLPNN) done by exact algorithm to find the maximum accuracy is NP-hard. Thus, we use the algorithm Back-Propagation who needs a starting point (weights and bias initials) to compute the training of the MLPNN. This research has developed and implemented a heuristic algorithm HeCI - Heuristic to Calculate Weights and Bias Initials - to compute the data to train the MLPNN and return the starting point for the Back-Propagation. HeCI uses Principal Component Analysis, Least Square Method, Probability Density Function of the Normal Gaussian Distribution, two strategic configurations, and partially controls the number of MLPNN training epochs. Experimentally, HeCI was used with Back-Propagation in MLPNN training to recognize patterns and solve data classification problems. Six case studies with datasets between Health, Business and Botany were used in the experiments. The methodology of this research uses Deductive analysis by the Experimental method with Quantitative approach and hypothesis tests: Test of Fridman with post Teste of Tukey HSD Post-hoc and Wilcoxon Test-M-W. The results of accuracy have increased significantly improving attested by evaluation of tests of hypotheses, inferringstatistical robustness of the result motivated by HeCI