Avaliação de Desempenho de Classificadores de Ciclos Hidrológicos em Reservatórios de Água na Região Amazônica

Jean Arouche FreireYuri SantosJefferson MoraisTerezinha Oliveira

This paper systematically evaluates classifiers in the prediction of hydrological cycles from the change of physico-chemical parameters and metals of water reservoir of the hydroelectric power plant Tucuruí. The methodology initially is to conduct an exploratory analysis of data in order to extract only the most relevant variables of the observed samples. The choice of the parameter values of the classifiers was made with automatic model selection. Results of applying Artificial Neural Networks, K-nearest neighbors, support vector machines and Random Forest techniques are presented. The results indicate that the Random Forest classifier showed the best performance with a percentage rating of 7.8 % of incorrect predictions. These values can be considered significant, since there is a great variability of physico-chemical parameters and metals in the hydrological cycles where are the sampling stations of study area.

http://www.lbd.dcc.ufmg.br/colecoes/wcama/2014/004.pdf

Caso o link acima esteja inválido, faça uma busca pelo texto completo na Web: Buscar na Web

BDBComp - Biblioteca Digital Brasileira de Computação
BDBComp
Parceria:
SBC
Avaliação de Desempenho de Classificadores de Ciclos Hidrológicos em Reservatórios de Água na Região Amazônica

Jean Arouche FreireYuri SantosJefferson MoraisTerezinha Oliveira

This paper systematically evaluates classifiers in the prediction of hydrological cycles from the change of physico-chemical parameters and metals of water reservoir of the hydroelectric power plant Tucuruí. The methodology initially is to conduct an exploratory analysis of data in order to extract only the most relevant variables of the observed samples. The choice of the parameter values of the classifiers was made with automatic model selection. Results of applying Artificial Neural Networks, K-nearest neighbors, support vector machines and Random Forest techniques are presented. The results indicate that the Random Forest classifier showed the best performance with a percentage rating of 7.8 % of incorrect predictions. These values can be considered significant, since there is a great variability of physico-chemical parameters and metals in the hydrological cycles where are the sampling stations of study area.

http://www.lbd.dcc.ufmg.br/colecoes/wcama/2014/004.pdf

Caso o link acima esteja inválido, faça uma busca pelo texto completo na Web: Buscar na Web

Biblioteca Digital Brasileira de Computação - Contato: bdbcomp@lbd.dcc.ufmg.br
     Mantida por:
LBD