The algorithm works as follows: for a network with i training examples of dimension d, and with a single output O. To compute the sensitivity along the domain, an x is taken for all the values of the domain and so PSIdx is calculated, then an importance values matrix is obtained, the rows will be the input variables and the columns, the importance of an input variable in a point x of the domain: PSIdx is calculated following the equation:Finally, a radial basis function neural network has been implemented with only all eucalyptus patterns; the method computes clusters for classification wood volume in a eucalyptus forest. It was performed an initial study using 150 patterns in training set and four input variables. All centers are stable in three points which show the three main clusters, and where the net has been possible to detect the three classes of tree (see Table 1). Main centers of RBF approximate real clusters in the three forest areas, following Table 3 shows the real clustering.The process carried out found three functions, which predict the volume of wood. After it should be check the results obtained by linear regression models, examining the likelihood ratios in the model, as well as, possible correlations between input variables. In this way, it is possible to compare the results obtained through the networks and regression model and the prediction error which is obtained with both models.
đang được dịch, vui lòng đợi..
