Data-Informed Tuning of Texture in Xanthan Gum-Based 3D-Printed Foods Using ANOVA and Machine Learning

3D food printing enables fabrication of personalized foods with tailored structures and textures; however, reproducible control of post-print texture remains challenging because formulation-process interactions are highly coupled and nonlinear. This study investigated the effects of extrusion speed, layer thickness, and nozzle temperature on the texture of xanthan gum-based printed foods using a hybrid framework combining statistical design of experiments and supervised machine learning (ML). A 3³ full-factorial design (27 parameter combinations) was implemented on an extrusion-based food printer, and hardness, cohesiveness, and chewiness were evaluated using texture profile analysis. Analysis of variance (ANOVA) and Taguchi nominal-the-best analysis revealed that extrusion speed mainly influenced hardness and chewiness, whereas layer thickness predominantly affected cohesiveness. Six regression-based ML models were evaluated, among which Random Forest (RF) showed the most favorable overall predictive performance within the studied dataset. Model-based evaluation of parameter combinations identified 20 mm/s extrusion speed, 0.3 mm layer thickness, and 90°C nozzle temperature as a balanced operating condition for the target texture ranges. Confirmation experiments showed only small deviations between predicted and measured responses. An additional intermediate "unseen" condition tested on a related screw-based printer also showed reasonable agreement with RF predictions, indicating local interpolation capability within the examined parameter range. However, because external validation was limited to a single intermediate condition on a related printer configuration, the findings should be interpreted within the studied process window rather than as broad hardware-independent generalization. PRACTICAL APPLICATIONS: The proposed ANOVA-ML workflow provides a systematic strategy for tuning extrusion parameters in XG-based 3DFP to achieve more consistent texture while reducing trial-and-error experimentation within the studied process window.