基于监督抽检数据的肉类食品安全风险分析及预测

为通过数据挖掘预测食品安全风险、隐患和趋势，进行预警和快速反应，基于2015—2017年国家肉类食品监督抽检的18 378 批次样品数据，分析我国肉与肉制品主要安全现状与风险种类，并基于检测指标及属性运用BP（back propagation）神经网络方法构建以抽样省份、产品类型、产地、生产日期、年份、是否大型企业6大属性指标为输入层、包含2 个隐藏层、以是否合格为输出层的肉类食品安全神经网络预测模型。结果表明：经数据准备、模型生成、数据训练和验证及参数优化，得到的3 层BP神经网络预警模型总体百分比矫正为96.2%；对于合格样本，判定正确的概率为96.5%，错判概率为3.5%，预测模型具有较好的参考和应用价值。基于BP神经网络的食品安全预警方法能够对输入样本进行有效预测，为食品安全风险研判和风险预警提供技术支撑。

Analysis and Prediction of Meat Product Safety Based on Supervision and Sampling Data

In order to predict food safety risks, hazards and trends through data mining for the purpose of early warning and rapid response, we collected 18 378 batches of supervision and sampling data on meat products in the period from 2015 to 2017 from the China Food and Drug Administration to analyze the current status of meat and meat product safety and the types of risks, and we further constructed a meat safety prediction model by back propagation (BP) neural network with two hidden layers based on the indices and attributes tested using sampling province, product type, geographical origin, date of production, year, and whether the manufacture is a large-size company as input variables and using whether products are qualified as output layer. The overall fitting accuracy of the three-layer neural network prediction model obtained after data preparation, model generation, data training and validation, and parameter optimization was 96.2%. For qualified samples, the probability of correct judgment was 96.5% and the probability of misjudgment was 3.5%. The model may serve as a reference and have application potential. The results show that the food safety early warning method based on BP neural network can effectively predict input samples and therefore can provide technical support for food safety risk analysis and early warning.