Identification and Ranking of Key Formulation Factors Affecting Sensory Texture in Bakery Products Using Machine Learning-Based Sensitivity Analysis

This study aimed to identify and rank key formulation factors influencing sensory texture in bakery products using interpretable machine learning. A dataset of ۲۸۷ formulations (bread, cake, biscuit) from peer-reviewed literature (۲۰۱۸-۲۰۲۴) was analyzed, with sensory scores (۱-۵) as the target variable and ingredient percentages as inputs. An XGBoost model coupled with SHAP analysis revealed that flour-to-water ratio was the most influential factor (mean |SHAP| = ۰.۴۱), followed by fat (۰.۳۳) and yeast (۰.۲۸), while sugar and salt had negligible effects. High yeast (>۲.۵%) with low water (<۴۵%) produced dense, sour textures due to gluten disruption and over-acidification, whereas optimal hydration (flour-to-water ratio ۱.۴-۱.۶) yielded springy, cohesive crumb. This work provides the first SHAP-based, data-driven framework for evidence-based formulation in bakery science, enabling targeted optimization to enhance quality, reduce waste, and replace trial-and-error practices in the food industry.

کلیدواژه ها:

Hydration Optimization ، Gluten Network ، SHAP ، XGBoost ، Food Science