Classificacąõ da capacidade produtiva de sítios florestais utilizando máquina de vetor de suporte e rede neural artificial

Abstract

Researchers in forest measurement have often included in their studies the use of computational intelligence (CI) techniques for modeling by being able to manipulate a large data set and create robust models. Among these techniques stands out Artificial Neural Network (ANN) and the latest Support Vector Machine (SVM). Therefore this study aimed to evaluate the use of these techniques (ANN and SVM) in site classification including some characteristics of soil, management and forest, comparing their results with those obtained by the guide curve method. It was concluded that CI techniques evaluated are able to classify sites satisfactorily since the appropriate variables are used; the combination of variables "soil type", "planting spacing", "age" and "dominant height" was sufficient to classify the sites; the ANN is better than SVM to site indexing; the inclusion of many low significance variables can be either detrimental or indifferent to the techniques performances.