Anadolu karaçamı [Pinus nigra J.F. Arnold subsp. pallasiana (Lamb.) Holmboe] meşcereleri (Bursa-Kestel) için çap-boy ilişkilerinin yapay sinir ağları ile modellenmesi

In this study, it is aiming to evaluate the usability of Artificial Neural Network Model to predict dbh and height relations for Kestel Forests located in Marmara region of our country. For this purpose, 500 Artificial Neural Network Model models with 5 activation functions and 100 neuron number alternatives were trained by using the height of trees measured in 110 sample areas and the height predictions were obtained. In comparisons based on some criteria such as r, Absolute Average Error (AAE), max. Absolute Average Error (max. AAE), Root Mean Squared Error (RMSE), Percent Root Mean Squared Error, Bias, Bias%, R2, Akaike Information Criteria (AIC) and Bayesian Information Criteria (BIC) , the Artificial Neural Network Model including activation function alternative (A2) with logistic sigmoid (log-sig) at connection point between the input layer and the hidden layer with the hyperbolic tangent sigmoid (tan-sig) at connection point between the hidden layer and the output layer and 75 neurons gave the best predictive results. For this the best predictive ANN model, r, AAE, max. AAE, RMSE, RMSE%, Bias, Bias%, R2, AIC and BIC were calculated as 0.9317, 0.9183, 5.1449, 1.2660, 10.4356, -0.0229, -0.1886, 0.8732, 197.3114 and 211.4062, respectively. According to Schnutte (1981) 's regression model, the best predictive ANN model gave improvements in the rates of AAE, RMSE, RMSE%, Bias, Bias%, R2, AIC and BIC of 4.518%, 12.296%, 35.689%, 22.929%, 22.929%, 51.703%, 49.978% -11.016%, respectively. When all these criterion values are evaluated, it can be concluded that ANN models can be used successfully in the height predictions.