Application of Artificial Neural Network in Predicting the Electrical Conductivity of Recombined Milk

In this study, the moisture content of kiwifruit in vacuum dryer was predicted using artificial neural networks (ANN) method. The protein (1, 2, 3 and 4%), lactose (4, 6, 8 and 10%), fat (3 and 6%) and temperature (50, 55, 60 and 65ºC) were considered as the independent input parameters and electrical conductivity of recombined milk as the dependent parameter. Experimental data obtained from electrical conductivity meter, were used for training and testing the network. In order to develop neural network firstly experimental data were randomly divided into three sets of training (70%), validating (15%) and testing model (15%).  In order to develop ANN models, we used multilayer perceptron with back propagation with momentum algorithm. MLP models trained as two, three and four layers. The total number of hidden layers and the number of neurons in each hidden layer were chosen by trial and error. The best training algorithm was LM with the least MSE value. The highest coefficient of determination (R2) and lowest mean squared error (MSE) were considered as the criterion for selecting the best network. The network having three layers with a topology of 4-4-1 had the best results in predicting the electrical conductivity of recombined milk. This network has two hidden layers with 8 neurons in the first hidden layer and 5 neurons in the second hidden layer.  For this network, R2 and MSE were 0.992 and 0.011, respectively. These results can be used in milk processing factories. The correlation between the predicted and experimental values in the optimal topologies was higher than 99%.