Abstract: Abstract: Nutraceuticals are herbal and food products with high nutritional values, which are commonly used to treat or prevent diseases. In personalized nutrition, there is an inter-individual response to nutraceuticals intervention, due to the fact that a sub population can benefit more than others. The traditional production of dietary supplements or functional food has very limited flexibility of fabrication to achieve personalized customization with nutrients or functional factors. The oral dosage forms can inevitably lead to insufficient or excessive intake. A type of 3D printing technology, fused deposition modeling (FDM) has been selected to create complex and accurate shapes using food-derived thermoplastic materials. Particularly, polylactic acid (PLA) has been successfully used to produce the scaffold of oral capsule. In this study, a 9-channel PLA capsule scaffold structure and Box-Behnken design were used to explore the effects of printing temperature, nozzle diameter, and printing speed on the FDM 3D printing accuracy. Meanwhile, the printing channel area error (MSE) was used as the response index during the optimization. A 3 × 4 × 1 three-layer neural network (NN) model was established. The coefficients of determination (R2) were above 0.998, indicating an excellent fitness between the actual and predicted MSE in the NN model. The nozzle diameter was the most influential factor. The MSE remained almost unchanged with the increase of printing temperature and speed at the nozzle diameter of 0.3 mm, where the MSE of the capsule channel area was less than 10%. In terms of energy consumption and efficiency, a low level of MSE was obtained with a lower temperature and faster speed. No significant difference was observed between the prediction values of the NN model and the experimental ones. The results indicated that the NN model can be used to predict the process of FDM 3D printing. An optimal combination of parameters was achieved, where the nozzle diameter was 0.3 mm, and the printing speed ranged from 25 mm/s to 30 mm/s, while the printing temperature ranged from 173 ℃ to 180 ℃ and the MSE ranged between 1%-2%. In this case, the printing accuracy of the capsule scaffold structure was better with the lower energy consumption and higher printing efficiency. A scanning electron microscope was used to characterize the internal microstructure and monitor the wiredrawing phenomena of the capsule scaffold. There was an uniform microstructure of the internal channel. The bonding between printing layers was compact when the printing temperature was fixed at 176 ℃ with the nozzle diameter of 0.3 mm and printing speed of 25 mm/s. An obvious wiredrawing was observed in the capsule scaffold during the high printing temperature or large nozzle diameter. The contraction of PLA was greater as the increase in the slice layers at the small diameter of the nozzle, leading to the loose internal structure and the poor adhesion between the printed layers. The optimized parameters were obtained in the NN model to print the delicate internal microstructure of the capsule scaffold. These findings can provide a theoretical basis for the personalized production of nutraceuticals, thereby realizing the customization and controlled release of minerals, vitamins, or functional factors.