A COMPARATIVE STUDY OF ARTIFICIAL NEURAL NETWORKS AND MULTIPLE REGRESSION ANALYSIS FOR MODELING SKIDDING TIME


Caliskan E., Sevim Y.

APPLIED ECOLOGY AND ENVIRONMENTAL RESEARCH, cilt.17, ss.1741-1756, 2019 (SCI İndekslerine Giren Dergi) identifier identifier

  • Cilt numarası: 17 Konu: 2
  • Basım Tarihi: 2019
  • Doi Numarası: 10.15666/aeer/1702_17411756
  • Dergi Adı: APPLIED ECOLOGY AND ENVIRONMENTAL RESEARCH
  • Sayfa Sayıları: ss.1741-1756

Özet

One of the most important functions of forests is providing raw material for wood. Timber extraction is among the most technically demanding, expensive and time consuming activities for wood raw material production. The analysis of timber extraction activities is complex and it is quite difficult to model them. Therefore, Artificial Neural Networks (ANN) frequently used as a modelling tool in the analysis of complex problems have been used to solve this issue. The aim of this study is to investigate the feasibility of ANN's including Multilayer Perceptron (MLP), Cascade Forward Back Propagation (CFBP) and compare the predictions for total time during log skidding operations stations in Eastern Black sea region (Giresun Forest District Directorates) of Turkey with those of the Multiple Regression Analysis (MRA). Moreover, standard times are calculated, and affective factors are determined, after which the effectiveness levels are evaluated at each working stage by way of timing determinations when skidder is used for timber extraction. The comparison of models were carried out by using the correlation coefficient, mean squared error, root mean square errors and mean absolute error. The comparison results indicate that MLP and CFBP models are better at predicting the total time during log skidding operations in comparison with the MRA model. These results have put forth that artificial neural networks have a greater prediction in comparison with multiple regression analysis for predicting the skidding time in timber extraction operations and that less erroneous results are obtained. It is observed that artificial neural networks can be preferred in cases for which the multiple regression analysis predictions have not been met and the analysis cannot be performed.