Interval Prediction of the Safety Risk of Soy Sauce and Pot-Roast Meat Products Based on WPD-ARIMA-GARCH Model

Abstract

In view of the drawback of traditional deterministic prediction that it cannot provide uncertainty information, this study proposed a prediction model that integrates point estimation and interval estimation, and innovatively applied it to the field of food safety risk pre-warning. In the point estimation, wavelet packet decomposition (WPD) was used to decompose the weekly risk level sequence and the autoregressive integrated moving average (ARIMA) model was used for prediction. In the interval estimation, the generalized autoregressive conditional heteroskedastic (GARCH) model was used to predict the residual. The WPD-ARIMA-GARCH model established in this study was applied to the safety risk prediction of soy sauce and pot-roast meat products from a certain region. The results showed that the safety risk of soy sauce and pot-roast meat products from this region was relatively high at the end of March and July in 2019, which was consistent with the actual situation. Meanwhile, in the risk prediction of soy sauce and pot-roast meat products from 10 different regions, the mean square error, mean absolute error, and mean absolute percentage error of the model were 1.626, 0.806, and 20.824, respectively, and the prediction interval normalized average and coverage width-based criterion values at the 90% confidence interval were both 0.024, which could cover all true values. Therefore, the model has high prediction accuracy and low error, is useful for risk control for the quality and safety of soy sauce and pot-roast meat products, and provide technical support for daily food safety supervision.

Keywords

• marinated meat products; wavelet packet decomposition; autoregressive integrated moving average model; generalized autoregressive conditional heteroskedastic model; interval estimation