Abstract BACKGROUND Tegillarca granosa is prone to spoilage and deterioration during storage due to the action of microorganisms and enzymes. The traditional shelf‐life prediction methods have problems such as strong destructiveness, long time consumption, complex operation and strict requirements for personnel. This study model constructed an intelligent prediction model of T. granosa based on a multilayer perceptron (MLP). RESULTS Under different storage temperatures (25, 4, −18 °C), the physicochemical indicators total volatile basic nitrogen (TVB‐N), thiobarbituric acid reactive substances (TBARS), total viable bacteria count (TVC), and sensory characteristics (color, electronic nose) of T. granosa all showed a deteriorating trend over time. Shelf‐life prediction model outputs the shelf life by inputting multidimensional variables such as TVB‐N, TBARS, color, electronic nose and TVC. The quality prediction models include three types: predicting the TVB‐N values and TBARS values by inputting storage temperature and days; predicting the TVB‐N value by inputting the response value of the electronic nose sensor. All the prediction models performed outstandingly, with the coefficient of determination ( R 2 ) remaining above 0.98, the mean absolute error controlled within 0.50, and the mean square error within 0.4. CONCLUSION The above results fully verified that the models have good prediction accuracy and stability. By applying machine learning technology, the shelf life and quality dynamic change patterns of aquatic products have been predicted rapidly, accurately, and non‐destructively, which has significant practical value for the quality control of aquatic products and the reduction of economic losses. © 2026 Society of Chemical Industry.
Yuan et al. (Fri,) studied this question.