Water sorption isotherms of dates: Modeling using GAB equation and artificial neural network approaches

Water sorption isotherms at 15, 25 and 45 degrees C were determined for two date varieties. Water sorption modeling was carried out using the five-par ameter Guggenheim-Anderson-de Boer (GAB) equation, a modified-GAB equation and a novel artificial neural network (ANN) approach. Modeling using the GA B equations used physical data as input, while the ANN approach used both p hysical and chemical compositional data. The five-parameter GAB equation ha d a lower mean relative error (approximately 7%) than the modified-GAB equa tion (approximately 16%), in predicting equilibrium moisture content (EMC). The effects of temperature on the water sorption isotherms were not eviden t with the five-parameter GAB equation. Although the temperature effects on water sorption isotherms were evident with the modified GAB equation, the overall error was very high. Neither GAB equation could predict water sorpt ion isotherm crossing, an effect observed in the experimental data. An ANN model, optimized by trial and error was superior to both GAB equations. It could predict EMC with a mean relative error. of 4.31% and standard en or o f moisture content of 1.36 g/kg. The correlation coefficients (r(2)) of the relationships between the actual and predicted values of equilibrium moist ure content and date varieties obtained by the ANN were 0.9978 and 0.9999 r espectively. The ANN model was able to capture water sorption isotherm cros sing due to temperature effects. Water activity and chemical compositional data, however had more impact upon the water sorption isotherms than temper ature.