Feature selection by genetic algorithm in nonlinear taper model

Tree stem profile results from a complex structure of shapes and dimensions determined by ecological processes within the forest. However, the feature selection in the development of taper models has been underinvestigated to date. We propose a genetic algorithm (GA) to assess factors that affect the stem taper and volume of Eucalyptus urograndis trees at different ages (2, 7, and 14 years) in Brazil. A total of 213 sample trees were measured in diameter and height along the stem, crown width, crown base height, crown length, and crown ratio. These variables and the stand age were supplied to the GA that selects variables, replacing those of Kozak’s 2004 model. The performance of models was evaluated using error statistics and residual plots. The GA model was efficient in predicting diameters and volumes, mainly by increasing the accuracy of the estimates in the extreme portions of the trees. This was attributed to the selection of morphometric variables as predictors of stem taper and volume, making them understandable in ecological terms. We highlight GA as a robust tool, since it incorporated the morphometric variables in Kozak’s model that contribute to the estimates.