Tunning machine learning algorithms for forestry modeling: a case study in the height-diameter relationship

Abstract

In the present study, four machine learning algorithms were applied in the task of modeling the height-diameter relationship of Pinus taeda L. stands at different ages. Hundreds of parameter combinations were tested for the k-nearest neighbors, random forests, support vector machines, and artificial neural networks algorithms. In order to select the best model for each algorithm, the grid search and the k-fold cross validation methods were applied. The selected models were used to predict the total height of individuals in an independent data set, and the results were compared to those obtained by linear regression models. The machine learning models presented similar statistical indicators to the linear regression models. However, they had less biased dispersion of residues, especially in the stratified analysis by age. The support vector machine and the artificial neural network were the most satisfactory models in precision and dispersion of residues.

Keywords:
Artificial intelligence; Grid search; Artificial neural networks; Cross validation