Número de indivíduos do algoritmo neat para estimativa do volume comercial da caatinga arbórea

Abstract

The Arboreous Caatinga is a vegetation typology of the Caatinga biome and the knowledge of its volumetric stock is fundamental in the elaboration of management plans. The objective of this work was to define the minimum number of individuals to obtain a satisfactory network to estimate the commercial volume of the Arboreous Caatinga. Commercial volume data with bark of 300 individuals were obtained to compose the data base for processing with Artificial Neural Networks using the genetic algorithm NEAT. The processing consisted of two phases, in which the first contained networks with 10 individuals up to 100 in intervals of 10 individuals and the second, networks of 100 individuals up to 1000 in intervals of 100 individuals. An analysis of the correlation and root square of mean error for each phase and analysis of variance of a one way and Scott-Knott averages test were performed to compare the different configurations that generated satisfactory results. Networks with a number of individuals of less than 300 were discarded and from 400 individuals with initial connection density of 0.1, presented the highest correlations and the lowest roots square of the mean error in the training of the networks, this being the minimum number recommended population size for the configuration of trained networks with the NEAT algorithm to model the commercial volume in Arboreous Caatinga.