Comparison of adaptive neuro-fuzzy inference system and artificial neural networks (MLP and RBF) for estimation of oxidation parameters of soybean oil added with curcumin

A analysis of soybean oil oxidation in presence of different concentrations of active substrate of turmeric rhizome (curcumin) (0.012, 0.016 and 0.02 %) at 25 and 55 degrees C based on oxidation parameters including peroxide value (PV), acid value (AV) and iodine value (IV) at specific time interval, was performed. Adaptive neuro-fuzzy inference system (ANFIS) and multilayer perceptron (MLP) and radial basis function (RBF) functions of artificial neural network (ANN) with three inputs (temperature and concentration, time of sampling) and three outputs (PV, AV and IV) were used for the construction of models that could predict the oxidation parameters and were compared to multiple linear regression (MLR). It was shown that the ANFIS model (R-2 = 0.98, 0.85 and 0.99 for PV, AV and IV, respectively) performed better compared to ANN (MLP and RBF) and MLR. Sensitivity analysis based on ANFIS model suggested the high sensitivity of oxidation parameters on temperature and concentrations of curcumin due to its high antioxidant activity to enhance soybean oil shelf life.