Multilayer perceptron neural networking for prediction of quality attributes of spray-dried vegetable oil powder

In this study, the multilayer perceptron (MLP) artificial neural networks (ANN) method was used to predict the various physiochemical attributes based on spray drying conditions for microencapsulated synergistic vegetable oil blend. This article also presents comparative studies between an MLP ANN and response surface methodology (RSM) in the modelling and prediction of quality attributes of microencapsulated oil blend. The MLP ANN was trained using experimental data comprising of inlet temperature and feed rate as input parameters with a set of quality attributes, viz. microencapsulation efficiency, peroxide value, moisture content, bulk density, colour, hygroscopicity and porosity as output responses with one hidden layer of three units. A good relationship was established between measured and predicted values with MLP topology. The final selected ANN model was compared to the RSM model for its modelling and predictive abilities based on performance indices, viz. RMSE, MAE and R² for each output responses. The developed neural network was able to predict efficiently different physico-chemical parameters studied for the microencapsulated vegetable oil blend with a R² values ranging between 0.75 and 0.98. The overall relative error during training (0.75) and testing (0.55) obtained was also satisfactory. Thus, MLP neural networking can be regarded as an efficient tool for the investigation, approximation and prediction of the microencapsulated characteristics of the vegetable oil blend.