Abstract: In order to establish a food safety risk prediction model based on data mining, we monitored whether samples of beef and lamb kebabs sold in Beijing in 2019 are adulterated. A total of 200 samples were collected from 100 sales agencies via 10 different sales channels. Pork-, cattle-, sheep-, chicken- and duck-derived ingredients in these samples were detected by real-time polymerase chain reaction (PCR) to judge whether they were adulterated. Based on the detection indexes and the sample information, back propagation (BP) neural network algorithm was used to build the risk prediction model for adulteration of beef and lamb kebabs. The results showed that the reporting limit (Ct value) of meat from each of the animal species tested obtained with the quality control samples was 28.0. On this basis, the non-acceptance rate of the 200 samples tested was 17.5% (35/200), the non-acceptance rate of the beef and lamb kebabs being 14% (14/100) and 21% (21/100), respectively. Meat adulteration with pork and/or duck was prevalent. In addition, based on the above survey data, a three-layer BP neural network prediction model was constructed by sequential data preparation, model generation, data training and verification, and parameter optimization. This model was 95.7% accurate in predicting the unaccepted samples, which could meet the purpose of risk prediction. The established model will provide the basis for the prevention and control of food safety risk.

Key words: beef and lamb kebabs; adulteration; real-time polymerase chain reaction; back propagation; predictive model