基于多源异构数据的肉类食品安全风险评级模型研究

Research on risk rating model for meat food safety based on multi-source heterogeneous data

中文关键词:  肉类食品安全风险  随机森林  决策树  机器学习  数据治理

英文关键词:meat food safety risks  random forest  decision tree  machine learning  data governance

中文摘要:

      目的 基于多源异构数据构建肉类食品安全风险评级模型。方法 依据2020—2023年肉类监督抽查数据、食品召回数据、实验室管理系统数据和行政处罚数据等多源异构数据, 分析肉类食品的全链安全风险因素类别。通过数据治理手段, 使用基于决策树的随机森林算法构建肉类食品安全风险评级模型。结果 肉类食品安全风险评级模型的准确率达到90%以上, 与实际情况基本吻合, 基于模型分析发现食品添加剂和微生物指标的风险程度要高于其他类别, 其中山梨酸、亚硝酸盐和胭脂红这3类食品添加剂具有较高风险, 菌落总数在微生物指标中具有较高风险, 在流通环节的菌落总数的不合格率要高于生产和餐饮环节。结论 基于随机森林的肉类食品安全风险评级模型可以推断肉类安全风险因素的风险程度, 该模型可以为政府和监管机构提供风险管理方面的数据支持, 并为监督抽检方向提供建议。

英文摘要:

      Objective To construct a meat food safety risk rating model based on multi-source heterogeneous data. Methods Based on multi-source heterogeneous data such as meat supervision sampling data, food recall data, laboratory management system data, and administrative penalty data from the years 2020 to 2023, the risk factors for the entire chain of meat food safety were analyzed. Using data governance methods, a meat food safety risk rating model was constructed using a decision tree-based random forest algorithm. Results The accuracy of the meat food safety risk rating model reached over 90% and was generally consistent with the actual situation. Based on the analysis of the model, it was found that the risk levels of food additives and microbial indicators were higher than other categories. Among them, 3 types of food additives, namely sorbic acid, nitrite and carmine, posed higher risks. The total colony count posed a higher risk among microbial indicators, and the non-compliance rate of total colony count in the distribution link was higher than in the production and catering links. Conclusion The random forest-based meat food safety risk rating model can infer the risk levels of meat safety factors. This model can provide data support for risk management to government and regulatory agencies, and offer recommendations for the direction of supervision and sampling.