Performance investigation of arrangement type of perceptron neural network to predict mass transfer kinetic of daikon ultrasound-osmotic dehydration

In this research, ultrasound-osmotic dehydration kinetic of daikon was predicted by different activation function of neural network such as logarithm sigmoid and tangent hyperbolic. Solid gain and water loss were selected as outputs and immersion time, concentration of osmotic solution, pretreatment type and moisture content were chosen as inputs. In this study, in other to achieved best result in predicting ultrasound-osmotic parameters of daikon was used different arrangement of the two types of neural network. The result showed that using arrangement of type II network with tangent hyperbolic activation function can be predicted daikon ultrasound-osmotic dehydration with higher performance than type I network arrangement. Best configuration of neural network was obtained with 17 neuron per hidden layer. This network was able to predicted solid gain and water loss with R2 values 0.996 and 0.993. This innovative technique could be successfully applied for quantitative monitoring of daikon quality changes during ultrasound-osmotic dehydration process.