Possibility of using neural networks for moisture ratio prediction in dried potatoes by means of different drying methods and evaluating physicochemical properties

Abstract

Potato cubes were dried by different drying methods.  After the end of drying process, the experimental data were first fitted to the four well-known drying models.  The results indicated that the logarithmic and page models performed better compared with the other models.  Also, in this study neural networks were used in order to possibly predict dried potato moisture ratio (y), based on three input variables of drying time, drying temperature and different methods.  The results revealed that, logsig function based on 10 neurons in the first hidden layer could perform as the best goodness configuration for predicting the moisture ratio.  The comparison of the obtained results of ANNs and classical modeling indicated that, the neural networks have a higher capability for predicting moisture ratio (R² values 0.9972 and 0.996, respectively) compared with classical modeling.  Furthermore, the physicochemical properties of dried potato such as shrinkage, starch gelatinization temperature, color change, etc. were also determined.

Keywords: extraction kinetics, time constant, extract concentration, extraction yield, medicinal herbs