Visual and Textural Cues Dominate Flavor in Consumer Choice of Roasted Almonds: An Explainable Artificial Intelligence Decoding

Roasting constitutes a pivotal process that influences the ultimate quality of almonds. However, directly linking the changes in the appearance, texture, and flavor to consumer perception remains a major challenge. This study applies an integrated analytical framework combining real-time thermodynamic monitoring, volatile profiling, and explainable artificial intelligence (XAI) to elucidate the relative contribution of visual and textural cues to consumer preference for roasted almonds. Our time-resolved mapping captured moisture-mediated texture evolution and Maillard-driven volatile formation, identifying 65 volatile organic compounds (VOCs) (corresponding to 84 signals including dimers and trimers in the spectral fingerprint). Relative odor activity value analysis quantified 10 key odorants, with (E)-2-pentenal and 3-methylbutanal serving as chemical markers for quality transitions. Hierarchical consumer clustering established three preference groups, which were accurately predicted (88.00% accuracy) by the random forest (RF) model. Shapley additive explanations (SHAP) analysis revealed that physicochemical attributes (color and hardness) contributed over 68.17% of the predictive power, decisively outweighing the combined influence of key odorants. This indicates that visual and textural properties serve as dominant, rapid heuristics in consumer decision-making, potentially preempting detailed flavor evaluation.