Walnut genetic improvement and orchard management would significantly benefit from accurate prediction of critical yield-related traits. In this study, an adaptive neuro-fuzzy interface system (ANFIS) was used to predict walnut kernel percentage and kernel quality. ANFIS uses principles of artificial neural network (ANN) learning as well as fuzzy principles. A total of 14 morphological characteristics of 100 walnut genotypes from Golestan province in Iran were used as model inputs. Correlation analysis and principal component analysis (PCA) were tested for their ability to reduce the model input numbers needed for accurate output. Eight features (four nut-related traits, four leaf characteristics) were revealed to be the most useful ANFIS input variables. Modeling data revealed ANFIS could predict walnut kernel percentage with a coefficient of determination (R2) of 99%. Accuracy in detection of kernel quality was also 99%. These results indicated that a combination of the fuzzy c-means (FCM) method with the hybrid training algorithm were the most useful when designing the ANFIS model. Therefore, ANFIS is a highly recommended tool for modeling walnut yield traits.

Keywords