Estimation of body weight of Sparus aurata with artificial neural network (MLP) and M۵P (nonlinear regression)–LR algorithms

In this study, morphometric features such as total length, standard length, and fork length obtained from a total of ۳۲۱ Sparus aurata samples, including ۱۶۴ females and ۱۵۷ males, captured between ۲۰۱۲ and ۲۰۱۳ from İskenderun Bay were used as input value, while weight was used as an output value. The Artificial Neural Network (MLP-Multi-L Layer Perceptron) as well as the M۵P algorithm and Linear Regression (LR) algorithm from version ۳.۷.۱۱ of the WEKA Program were applied. When coefficients of correlation were assessed, the MLP algorithm for males, females and the total were calculated as ۰.۹۶۸۶, ۰.۹۶۰۵ and ۰.۹۶۶۳, respectively; the M۵P algorithm for males, females and the total were calculated as ۰.۹۷۲۲, ۰.۹۵۹۶ and ۰.۹۷۳۵, respectively; and the LR Model for males, females and the total were calculated as ۰.۹۷۷۷, ۰.۹۴۹۸ and ۰.۹۴۷۳, respectively. With respect to the Mean Absolute Error (MAE) calculations, the MLP algorithm MAE values for males, females and the total were calculated as ۲.۹۴, ۲.۵۷ and ۲.۷۰۷۴, respectively; the M۵P algorithm MAE values for males, females and the total were calculated as ۲.۴۰۰, ۲.۶۴۱ and ۲.۱۵۷, respectively; and the LR Model MAE values for males, females and the total were calculated as ۳.۲۱۷, ۲.۸۱۱ and ۳.۱۱, respectively. It can also be concluded from the study that, in order to predict ANN interactions Nonlinear Regression model is more effective and has better performance than the conventional models.

Keywords:

Weka ۳.۷.۱۱ , Artificial Neural Network-MLP , M۵P , Sparus aurata , Morphometric feature , İskenderun Bay