Moisture content and water activity prediction of semi-finished cassava crackers from drying process with artificial neural network

This paper is concerned with a prediction of moisture content and water activity of semi-finished cassava crackers from a hot air drying process in a tray dryer. A Multilayer Feedforward Neural Network in the form of Nonlinear-Auto-Regressive with Exogenous input (MFNN-NARX) was proposed to predict the product quality from the dynamic drying process. The process was carried out at seven drying temperature settings of 50, 55, 60, 65, 70, 75, and 80 °C. The MFNN-NARX was composed of one hidden layer, three exogenous inputs (drying temperature, relative humidity, and sample temperature), and two state inputs and outputs (moisture content and water activity). A number of hidden neurons and transfer functions were investigated in this study. Based on our results, the best network was composed of nine hidden nodes and used a logarithmic sigmoid transfer function in the first layer. The mean squared error (MSE) and regression coefficient (r2) between the normalized predicted and experimental outputs from the best network were 0.0034 and 0.9910, respectively. A simulation test with a testing data set showed that MSE was low and r2 was close to 1. This result showed the good generalization of the developed model.