A data-driven intelligent framework for silver carp (Hypophthalmichthys molitrix) surimi gel properties prediction and quality improvement.

This study developed a data-driven framework linking silver carp surimi gelling properties prediction with process optimization using raw-material characteristics and controllable processing parameters. A dataset of 1571 processing records spanning regions, seasons, fish-size grades, and key processing conditions was established. Target-specific stacked ensemble models predicted gel strength, breaking strain, and whiteness, with test set R2 values of 0.95, 0.82, and 0.93, respectively. SHAP analysis identified salt-soluble protein content and surimi moisture content as the most influential predictors of gel strength; region and season were most influential for breaking strain, whereas whiteness was more closely associated with season and the moisture-protein balance. The models were coupled with desirability-based optimization to generate raw-material-specific processing recommendations. Independent validation using six newly collected raw-material lots representing three regions and two seasons showed that the optimized schemes increased mean gel strength by approximately 25-107% relative to the conventional empirical process, with no significant differences detected in breaking strain; whiteness responses varied with raw-material and seasonal conditions. Overall, these results show that incorporating raw-material properties into interpretable multi-target prediction can support batch-specific process recommendations, and the desirable quality control.