Abstract
The purpose of this paper was to estimate for site indices in a Japanese Cedar forest with the machine learning system C4.5. Data inputted to the machine learning system consisted of four topographical factors : effective relief, topographical exposure, enabled storage capacity and flow accumulation. The output data set up four patterns. Classes 3, 5 and 7 were calculated based on the maximum-minimum value of the observed tree height, and Class 4 was calculated using site indices. Estimations by the machine learning system were compared with estimated results by both a multiple regression analysis model and a neural network model. As a result, the best classification accuracy and the best estimation accuracy for unseen cases were 87.5% and 66.6%, respectively. The machine learning system was superior to the multiple regression analysis model, though its estimation accuracy was slightly inferior to that of the neural network model. It was possible to construct a model with comparatively few rules ; the number of production rules was ≤7.
