Comparison between mathematical models and artificial neural networks for prediction of sorption isotherm in rough rice

Equilibrium moisture content data for long grain rough rice (Oryza Saliva, cv. Binam) were obtained by equilibrating rough rice samples at different equilibrium relative humidity (ERH) and temperatures. Although conventional mathematical models are able to predict EMC with high accuracy, such models can be competed and replaced with artificial neural networks (ANNs) method which is a simple mathematical model of human brain performance. Modified models of Chung-Prost, Halsey, Henderson, swill as well as GAB were used as mathematical models to fit the data. One of the multi layer perceptron (MLP) neural network types, called Feed Forward Back Propagation (FFBP), was used in this work. Training algorithm of Levenberg-Marquardt (LM) was also applied. The range of temperature was 0-35 with 5 degrees C intervals and relative humidity was 19.75-94.21%. The best results for mathematical model belonged to the Chung-Prost model with average R-2 = 0.9861 and mean relative error = 4.76%, and the best one for FFBP neural network with training algorithm of LM was appertained to the topology of 2-4-3-1 and threshold functions order of TANSIG-TANSIG-PURELIN. By the use of this optimized network, R-2 = 0.9958 and mean relative error = 3.56% were determined. These results show that mathematical models can be replaced with the ANNs for the prediction of EMC in the Binam variety of rough rice.