Comparison between artificial neural network (ANN) and regression analysis in tree felling time estimation

• Published in: Taḥqīqāt-i jangal va ṣanubar-i Īrān Vol. 20; no. 4; pp. 607 - 595
• Main Authors: Hadi Bayati, Akbar Najafi, Parviz Abdolmaleki
• Format: Journal Article
• Language: Persian
• Published: Research Institute of Forests and Rangelands of Iran 01.12.2012
• Subjects: Forest engineering; Forest harvesting; Multi Layer Perceptron; radial basis function; time study
• ISSN: 1735-0883; 2383-1146
• DOI: 10.22092/ijfpr.2012.107464

Tree felling is a most important one among the tree harvesting components. Production estimation of forest equipments is an important part of cost management in forestry operational units which is associated with reduction of the operating expenses. In other words, the high cost of investment in forest utilization, is a good reason for forest engineering research and modeling time. Many techniques such as regression, fuzzy logic, neural networks, etc. are utilized to estimate trees felling time. They make a logical connection between the tree felling time and the independent variables and could be used to predict the tree felling time for the future operations. In this study the regression analysis, two neural network models, multi-layer perceptron (MLP) and radial basis function (RBF) were used to predict the trees felling time in the cutting operations of the Neka Choob Co. In order to collect the felling time data, the time continuous study method was applied. For this purpose, 84 trees were selected from the marked stands and the net felling time was estimated, using the Multi Layer Perceptron and Radial Basis Function and also by the common method of linear regression analysis. The results showed that the Radial Basis Function network provided more accurate results in estimating the net tree felling time than the MLP neural network. Comparing the evaluation criteria of ANN with the stepwise regression methods,  showed that MLP and RBF neural networks had RMSE value of 0.94 and 0.81, respectively whereas the RMSE value of the regression model was 1.15.