Classification of Hybrid Chestnut Cultivars (Castanea sativa) Registered in Türkiye with Artificial Neural Networks, Based on Some Physical Properties of Their Nuts

Abstract

Understanding how the classification and identification of biological species can evaluate improvements in newly developed cultivars, including chestnuts (Castanea sativa), is crucial for product processing and equipment design. To evaluate this, in the present study artificial neural networks (ANNs) were used to characterize four hybrid chestnut cultivars (Macit 55, Akyuz, and Ali Nihat registered in Turkiye and Bouche de Betizac registered in France). A backpropagation neural networks algorithm was used in the ANN approach based on nine physical properties. These properties included shelled nut weight and volume, sphericity, geometric mean diameter, bulk density, surface area, true density, porosity, and length, which can be deemed for classifying the cultivars. The ANN model was composed of input (9), hidden (6-5), and output (1) layers. In the hidden layers and output layer, tansig transfer and linear transfer functions were used, respectively. The R2 value for the test and training data was 0.99999 (RMSE = 0.000083 and 0.0023, respectively). The relative error (epsilon) between the real values and the estimated values was 0.079%. In conclusion, the ANN approach is able to discriminate among Macit 55, Akyuz, Ali Nihat, and Bouche de Betizac accessions based on the values of R2 and epsilon.