基于集成学习的重大活动肉及肉制品安全风险分析及预测

Safety risk analysis and prediction of meat and meat products in major activities based on ensemble learning

中文关键词:  重大活动  肉及肉制品  集成学习  安全分析  风险预测

英文关键词:major activities  meat and meat products  ensemble learning  safety analysis  risk prediction

中文摘要:

      目的 建立集成学习Stacking模型, 对重大活动举办过程中的肉及肉制品进行合格安全分析与风险预测预警。方法 通过收集2015—2020年间国家市场监督管理总局食品日常监督管理抽检数据, 筛选出所有肉及肉制品相关数据, 选择“食品亚类”“规格”“生产时间”“生产企业类型”“生产省份”“是否异地运输”等字段信息作为肉及肉制品合格风险因子, 选取所有不合格数据5866条, 从所有合格数据中随机抽取10000条数据共15866条构成数据集, 按照3:1划分训练集和测试集, 搭建基学习器为K最近邻(K-nearest neighbor, KNN)、反向传播(back propagation, BP)神经网络、支持向量机(support vector machines, SVM), 元学习器为逻辑回归(logistics regression, LR)的Stacking预测模型, 进行训练预测与模型评估。结果 经过5次训练后, 模型的准确度为94.20%, 精确度为93.78%, 召回率为97.57%, F1参数为95.63%, 模型鲁棒性强, 可靠性高。结论 基于集成学习Stacking的肉及肉制品安全风险分析与预测预警模型整体性能良好, 可应用于重大活动举办过程中的食品安全风险分析预测, 并精确指导监督抽检与辅助决策情报研判。

英文摘要:

      Objective To establish an ensemble learning stacking model, conduct qualified safety analysis and risk prediction and early warning for meat and meat products during major events. Methods By collecting the sampling data of daily food supervision and management of the State Administration of Market Supervision and Administration from 2015 to 2020, all data related to meat and meat products were screened, and the field information such as “food category” “specification” “production time” “type of production enterprise” “production province” and “whether it is transported from other places” were selected as the qualified risk factors of meat and meat products. The 5866 pieces of unqualified data were selected, and 10000 pieces of data were randomly selected from all qualified data, with a total of 15866 pieces forming a data set, and the training set and testing set were divided according to 3:1, build a stacking prediction model with K-nearest neighbor (KNN), back propagation (BP) neural network and support vector machines (SVM) as the base learner and logistics regression (LR) as the meta learner for training prediction and model evaluation. Results After 5 training, the accuracy of the model was 94.20%, the precision was 93.78%, the recall rate was 97.57%, and the F1 arameters was 95.63%. The model had strong robustness and high reliability. Conclusion The qualified safety risk analysis, prediction and early warning model of meat and meat products based on ensemble learning stacking has good overall performance. It can be applied to the analysis and prediction of food safety risk in the process of major events, and accurately guide the supervision and sampling inspection and auxiliary decision-making information research and judgment.