Artificial neural network modelling of changes in physical and chemical properties of cocoa powder mixtures during agglomeration

An artificial neural network (ANN) which predicts the influence of agglomeration process parameters on physical and chemical properties of cocoa powder mixtures simultaneously, was developed. Cocoa powder mixtures were formulated with cocoa powders of different fat content (10-12 g/100 g and 16-18 g/100 g) and various sweeteners (carbohydrate sweeteners, sugar alcohols, intense sweeteners, bulking agents) and then subjected to agglomeration. For the design of ANN, agglomeration conditions (added water and agglomeration duration) and mixture composition (fat content, sweeteners content and bulking agent content) were used as input variables, and selected physical (Sauter diameter, bulk density, porosity, Chroma wettability and solubility) and chemical (total phenolic content and antioxidant capacity) properties as output variables. Based on the experimental data, agglomerated cocoa mixtures formulated with cocoa powder containing higher fat content (16-18 g/100 g) exhibited higher Sauter diameter, but poorer wettability and lower polyphenolic content and antioxidant capacity. The presented ANN model accurately predicts the effect of the five input parameters simultaneously on the output parameters (training R2 = 0.969; test R2 = 0.945; validation R2 = 0.934). Global sensitivity analysis revealed that the amount of water added during the agglomeration process influenced both physical and chemical properties of the agglomerated cocoa powder mixtures the most.