Tulīd va Farāvarī-i Maḥṣūlāt-i Zirā̒ī va Bāghī (Sep 2015)
Soybean Yield Prediction Using Adaptive Nero-Fuzzy Interface System (ANFIS)
Abstract
Productivity of rainfed crops may be predicted using the climatic parameters. Crop yield prediction has an important role in agricultural policies including determining the crop price. Well-known prediction methods are regression method and arterial neural networks. In this paper soybean yield is predicted using Adaptive Nero-Fuzzy Interface System (ANFIS) and 11 years of climatic data (1998-2009) in Gonbad-e-Kavous region of Golestan province, Iran. Mean weekly rainfall, mean weekly temperature, mean weekly relative humidity and mean weekly sun shine hours were ANFIS inputs and its output was soybean grain yield (kg/ha). Stepwise Regression for Feature selection from climatic data was done with the SPSS18 software and ANFIS was created, trained and tested with MATLAB R2011a software. Trained ANFIS has ‘constant’ membership function in output layer and ‘gaussmf’ membership function in input layer. Each input has 3 membership functions and each output has one membership function. Root Mean Square Error (RMSE) criterion was used to evaluate the performance of the ANFIS. The results showed that the proposed ANFIS with 21 rules has a prediction error (RMSE) of 102.170.