Response surface optimization of an artificial neural network for predicting the size of re-assembled casein micelles

An artificial neural network (ANN) was designed to predict the size of re-assembled micelles in casein solutions as influenced by pH of solution and ultrasonic treatment. A generalized feed-forward network consisted of five neurons in the input layer, one hidden layer and an output layer with one neuron optimized using response surface methodology (RSM). Number of hidden neurons, momentum coefficient and step size in the hidden layer, number of epochs and training runs were the variables optimized. A quadratic equation was applied to mean absolute error (MAE) of 52 artificial neural networks as the response. It was found that the first-order effect of epoch number is the most significant term in determination of MAE, followed by the interactive effect of epoch number and step size. Minimum response (MAE) was obtained by employing the following optimum conditions for the artificial neural network: hidden neurons number = 10, momentum coefficient = 0.6, step size = 0.34, epoch number = 6230 and training run = 1.