Exploring the best model for sorting Blood orange using ANFIS method

Abstract

OranOrange has abundant nutritional properties and is consumed worldwide.  Sorting oranges of different masses based on their physical traits could help reduce packaging and transportation cost.  The ‘Blood’ cultivar of Iranian oranges from Kermanshah province of Iran (7.03 °E 4.22 °N) was used in this study.  100 samples were randomly selected.  During the two-day experiment, all measurements were carried out inside the laboratory at mean temperature of 24°C.  In this study, some physical properties of ‘Blood’ orange were measured, such as length, width, thickness, volume, mass, mean value of geometric diameter, sphericity and projected area.  ANFIS and linear regression models were employed to predict the mass based on sphericity and mean of projected area inputs.  In ANFIS model, samples were divided into two sets, with 70% for training set and 30% for testing set.  The coefficient of determination (R²) for ANFIS and linear regression models were 0.983 and 0.927, respectively.  It is shown that the mass can be estimated based on ANFIS model better than linear regression model.

Keywords: linear regression, orange, packaging, physical properties, sorting