ANN modeling of the extruded products physical and technological properties

Abstract

This study used artificial neural network (ANN) modeling to predict and optimize the physical and technological properties of quinoa- enriched extruded products. A multilayer perceptron model was developed to estimate bulk density (BD), expansion index (EI), hardness (Har), number of fractures (NF), and crispiness work (Crw) as functions of screw speed (350–650 rpm) and quinoa addition (0–30%). The model demonstrated high predictive accuracy and successfully passed goodness-of-fit tests. Global sensitivity analysis revealed that screw speed was the dominant factor, while higher quinoa levels and screw speeds promoted expansion and reduced density and hardness. The ANN approach proved effective for modeling and optimization of extrusion conditions.