期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2020
卷号：98
期号：19
页码：3233-3241
出版社：Journal of Theoretical and Applied
摘要：Forest plays an important role in national growth as the forestry and logging activities contribute 5.6% to the Malaysia GDP of the agricultural sector in 2018. A precise value of tree volume estimation highly affects forest management and administration. The forest management and administration framework are designed based on the evaluation of the forest, including its current volume; therefore this strongly supports the need for a precise tree volume estimation. Tree volume can be expressed either in terms of the total cubic volume of a tree or in terms of the total cubic volume of an area. However, this paper is going to focus on the volume estimation technique for an individual tree. Analysis of the literature found that the commonly used method in estimating the tree volume is regression, however, growth in the information technology has driven the use of machine learning techniques. The state-of-the-art highlighted that machine learning not only has a high capability in developing a robust model but also able to overcome the regression analysis problem such as overfitting of the data. Numerous comparison studies on the application of machine learning in forest modelling can be found but there are discrepancies of analysis among scholars. Therefore, this paper will perform tree volume estimation by using regression and four machine learning techniques which are artificial neural network (ANN), epsilon-Support Vector Regression (ε-SVR), k-Nearest Neighbor (k-NN) and random forest (RF). The precision and accuracy of the volume model will be verified by using the root mean square error (RMSE) and standard deviation (SD). The result and analysis of this study seem to be consistent with other research which found that machine learning techniques perform better than regression as the ANN is the best modelling technique for dipterocarp and non-dipterocarp datasets while for all species dataset, ε-SVR records the highest accuracy.
