Application of data mining methods in estimation of hazelnut yield in orchards equipped with pressurized irrigation in Gilan province

Kiwi is one of the Iran's agricultural products, which is mainly exported to the world market. Data mining methods are well able to provide manufacturers with the necessary information in the field of product performance modeling. This research has investigated the efficiency of data mining methods of advanced Feedforward backdrop Neural Network, K-Nearest Neighbor, Genetic Programming and Multivariate Linear Regression in estimating kiwi yield in Gilan province using water and soil characteristics. 74 data series were obtained from the field measurement of water and soil information and yield of kiwi fruit in orchards equipped with pressure irrigation system in 2021-2022. Water and soil data including maximum daily evaporation and transpiration, soil electrical conductivity and soil reaction index, clay percentage, silt percentage, soil organic matter percentage, water electrical conductivity and water reaction index, and irrigation volume as model inputs and crop yield, model output selected. The results showed that the leading artificial neural network model has a better performance than the other three models due to the higher explanation coefficient statistics (0.96) and lower root mean square error (0.019). Also, Genetic Programming has correlation coefficient (0.89), root mean square error (0.033) and K-Nearest Neighbor method has correlation coefficient (0.88) root mean square error (0.059) and Multivariate Linear Regression method has coefficient The correlation was (0.58) and the root mean square error (0.093) which indicates the higher accuracy of the Genetic Programming method. Therefore, the advanced artificial neural network model can act as a powerful tool in the estimation of Kiwi performance.