Prediction of Equilibrium Moisture Content of Various Foods Using a Neural Network

A neural network was developed for learning the isotherm data and determined the relationship of equilibrium moisture content and relative humidity at various temperatures and for different food products. Data were taken from the literature for 53 food products. Neural network inputs were temperature, relative humidity or water activity, and food product number (1–53) or six binary elements (000001 to 110101) to express these food products. Food product input in binary numbers provided much better accuracy than using numerical numbers. The optimum neural network structure was three middle layers with 30 nodes in each layer with 0.9 values of both learning rate and momentum. The mean relative error was only 2%.