Studying the relationships between nutrients in pistachio leaves and its yield using hybrid GA-ANN model-based feature selection

Abstract

Sustainable and reliable management requires special attention to factors affecting crop yield. In the present study, a hybrid model of genetic algorithm and artificial neural network (GA-ANN) was employed to recognize the importance of nutrients in pistachio yield. One hundred seventy-five points in different pistachio orchards of Rafsanjan and Anar regions, Kerman province, the southeast of Iran, were identified and selected for leaf sampling and yield measurement. The concentration of phosphorus (P), potassium (K), iron (Fe), zinc (Zn), copper (Cu), manganese (Mn), calcium (Ca) and magnesium (Mg) was determined. The hybrid GA-ANN model was implemented in MATLAB software, after statistical analysis and multivariate regression modeling. The results showed that the correlation and linear multiple regression analysis could not justify the variations of pistachio yield in relation to leaves' nutrients concentration. The lowest error of the hybrid GA-ANN model was observed by five features including concentrations of K, Mg, Fe, Zn and Cu. Sensitivity analysis of ANN indicated that the highest relative importance for predicting pistachio yield was related to Cu (34.6%), K (28.2%) and Fe (26.1%). The GA-ANN model was able to solve complex and multi-dimensional problems. The accurate and careful interpretation of the results, obtained from this approach can provide a good insight for optimum farm management planning.