Abstract
In this research, the main goal is to develop and apply a robust adaptive fuzzy inference system (ANFIS) and Analysis of Variance (ANOVA) models to predict the mass transfer during osmotically treated zucchini samples as a heat-sensitive product. Maltose, fructose, lactose, and fructo-oligosaccharide (FoS), as osmotic agents at various concentrations and at constant temperature, were applied. In this study, the investigated concentrations were 30%, 40%, and 50% and the temperature was at 26 degrees C. Three performance outputs were considered: Dimensionless Moisture Content (DMC), Water Loss (WL), and Solid Gain (SG). Using the dataset obtained from the experiments, both ANFIS and ANOVA models were proposed to describe osmotic dehydration behavior. The inputs to the models are the time (min), the concentrations (%) and the osmotic agents. During the training and testing phases, the ANFIS showed significant modeling performances relative to the ANOVA in terms of the coefficient-of-determination (R-2) and the RMSE values. In particular, for the whole dataset, the RMSE values are improved using ANFIS compared to the ANOVA where they are decreased by 99.03%, 93.5% and 79.08%, for the DMC, SG and WL, respectively. On the other hand, the coefficient-of-determination showed significant increases for the ANFIS models over the ANOVA models for both training and testing phases. The comparison between the two modeling techniques revealed that the ANFIS is found more suitable for predicting all three performance outputs during the Osmotic Dehydration (OD) process of the zucchini food product. The results showed that the maltose, as an osmotic agent, with a concentration of 30% is the most suitable solution among the four osmotic agents that produced the best performance for the three performance outputs. In sum, the obtained results demonstrated the superiority of ANFIS modeling of the DMC, SG, and WL in comparison with ANOVA. Accordingly, the prediction from the ANFIS model has an excellent agreement with the experimental dataset. This confirms the accuracy of the ANFIS model.