Application of a combination model based on an error-correcting technique to predict quality changes of vacuum-packed bighead carp (Aristichthys nobilis) fillets

• Quality changes of fish fillets were modelled and predicted by combination model.
• Combination model included kinetic model and ANN.
• Results of kinetic model were revised successfully by ANN.
• Combination model performed better than single kinetic model.

A combination model based on an error-correcting technique was developed to predict changes in sensory scores, TAC, and K-values of vacuum-packed bighead carp fillets during storage at different temperatures (12, 9, 6, 3, and 0 °C). The combination model included a kinetic model and an artificial neuronal network (ANN). TAC, K-values, and sensory scores were modelled by zero-order kinetics, and residual errors generated were simulated by ANN. Then, error corrections obtained by ANN were used to revise results of kinetic models. Relative errors of kinetic models exceeded 10% on some days, by contrast, the proposed combination models performed better with relative errors all within 5%. Therefore, combination models were more satisfactory than single kinetic models.