机器学习在食品安全风险预警及抽检方案制订中的应用研究  

Application Research of Machine Learning in Food Safety Risk Early Warning and Sampling Inspection Program

作　　者：杨鸿雁 田英杰 Yang Hongyan;Tian Yingjie(School of Economics and Management,University of Chinese Academy of Sciences,Beijing 100190;CAS Research Center on Fictitious Economy&Data Science,Beijing 100190)

机构地区：[1]中国科学院大学经济与管理学院,北京100190 [2]中国科学院虚拟经济与数据科学研究中心,北京100190

出　　处：《管理评论》2022年第11期315-323,共9页Management Review

摘　　要：“问题导向”的食品安全监督抽检方案对加强食品安全风险管理具有极重要作用。现有关于食品抽检数据的研究较少考虑到不合格样本极少的不均衡特性。本文基于新疆2015—2017年的食品监督抽检数据,将不均衡问题解决思路引入对食品安全风险预警模型的构建研究中,分别构建多种采样方法、代价敏感方法与SVM、随机森林的组合算法模型。经分析比较,在解决食品抽检不均衡数据的问题方面,采样方法逊于代价敏感方法。关于风险预警模型的构建,代价敏感+SVM的组合性能虽稍优于代价敏感+随机森林组合,但前者的运行时间远多于后者。实证结果表明,基于不均衡数据解决思路构建的组合算法模型能够有效提升食品安全风险预警效果,为制订科学合理的食品安全监督抽检方案、提升食品安全风险管理效能提供决策支持。The problem-oriented sampling inspection program of food safety supervision plays an extremely important role in strengthening food safety risk management.Existing researches on food sampling data rarely consider the unbalanced characteristics of unqualified samples.Based on the food inspection and sampling data of Xinjiang from 2015 to 2017,this paper introduces the solution of unbalanced problems into the research on the construction of early warning models of food safety risks,and constructs a combined algorithm model of multiple sampling methods,cost-sensitive methods and SVM,and random forest.Our analysis and comparison show that sampling methods are inferior to cost-sensitive method in solving the problem of unbalanced data in food sampling.Regarding the construction of the risk warning model,although the cost-sensitive+SVM combination performance is slightly better than the cost-sensitive+random forest combination,the running time of the former is much longer than the latter.The empirical results show that the combined algorithm model based on imbalanced data solution can effectively improve the early warning effect of food safety risks,and provide decision support for formulating scientific and reasonable sampling inspection program of food safety supervision and improving the effectiveness of food safety risk management.

关 键 词：监督抽检 不均衡问题 组合算法模型 风险预警 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] F203[自动化与计算机技术—控制科学与工程]