In this study, the laboratory data of an experimental paper on separating the food process by ultrafiltration was used and entered as the input experimental data for the MLP multi-layer neural network modeling. The operating parameters were pressure, time, frequency, and flux. According to the created results and the regression coefficient R2 and MSE values evaluation, the Levenberg-Marquart training function has been selected as the optimal training function. Also, five important and widely used transfer functions named tansig, purelin, radbas, satlin, and logsig were evaluated. Among them, the tansig transfer function with R2=0.99783 and the number of neurons 16 was selected as the optimal transfer function. According to the examination of R2 and MSE values, the 70-15-15 combination has been selected as the optimal combination for the number of input data to the training, testing, and validation section.
