AI-driven modelling and experimental analysis of oil concentration impact on mayonnaise rheology for innovative food design

Abstract

This study investigates the influence of oil concentration on the rheological behavior of mayonnaise by integrating experimental methods with machine learning-based predictive modelling. Self-made mayonnaise samples prepared with varying oil content and a commercial sample were analyzed through comprehensive rheological testing. Results demonstrated that increased oil content enhanced viscosity, yield stress, and viscoelastic structure. A sample with 70 % oil content exhibited rheological properties and optimal thixotropic recovery (∼70 %) most comparable to the commercial product. The Herschel-Bulkley model provided a better fit than the Power Law for flow behavior characterization. Machine learning models were trained to predict viscosity from rheological parameters, with XGBoost algorithm achieving the highest prediction accuracy (R² = 0.966), outperforming Gradient Boosting, Random Forest, and other models. Feature sensitivity and SHAP analysis identified shear rate and oil concentration as the dominant factors influencing viscosity. Overall, the study presents a novel, data-driven methodology for characterizing and modelling emulsified food rheology. The findings offer valuable insights for formulation, process optimization, and demonstrate the potential of machine learning to support efficient, scalable food product development.