基于机器学习算法构建酱卤肉货架期预测模型  

Constructing A Shelf⁃life Prediction Model for Marinated Meat Products Based on Machine Learning Algorithms

作　　者：张慧娟 黄千里 徐宝才 ZHANG Huijuan;HUANG Qianli;XU Baocai(School of Bioengineering and Food Engineering,Hefei University of Technology,Hefei 230009,Anhui,China)

机构地区：[1]合肥工业大学食品与生物工程学院,安徽合肥230009

出　　处：《食品研究与开发》2024年第9期143-149,192,共8页Food Research and Development

摘　　要：将机器学习算法和文本挖掘融入酱卤肉制品货架期预测中,基于对文献数据库中酱卤肉制品的货架期及其影响因素(包装方式、储藏方式、保鲜剂和二次杀菌)进行收集,构建原始数据集;通过比较多种编码方法(JamesStein、BaseNEncoder、TargetEncoder、OrdinalEncoder、PolynomialEncoder),选择效果较好的JamesStein编码作为分类型特征变量的编码方式。通过比较多种机器学习算法(包括随机森林算法、K最近邻算法、逻辑回归、XGboost和多层感知机分类器),结果显示最优模型为随机森林算法[其准确度为0.95、精确度为0.97、曲线下面积(area under curve,AUC)值为0.99,F1-score 0.91]。通过对酱牛肉和盐水鸭的实际样品测试分析,发现该模型在预测不同酱卤肉制品的货架期方面均具有较高的准确性。此外,该文从另一个角度验证储藏温度、包装方式、保鲜剂和二次杀菌等因素对酱卤肉制品货架期的显著影响。This study integrated machine learning methods and text mining into the shelf⁃life prediction re⁃search of marinated meat products.an original dataset was constructed based on the collection of shelf⁃life and influencing factors(packaging methods,storage methods,preservatives,and secondary sterilization)of mari⁃nated meat products in literature databases.By comparing various encoding methods(including JamesStein、BaseNEncoder、TargetEncoder、OrdinalEncoder、PolynomialEncoder),the James⁃Stein encoding was selected as the encoding method for categorical feature variables with better performance.Subsequently,through com⁃paring various machine learning algorithms(including RandomForest,K⁃nearest neighbors,LogisticRegres⁃sion,XGboost,and multi⁃layer perceptron classifier),the optimal model was found to be a random forest(with an accuracy of 0.95,precision of 0.97,AUC value of 0.99,and F1⁃score of 0.91).Testing and analysis of actual samples of marinated beef and salted duck confirmed the high accuracy of the model in predicting the shelf⁃life of different marinated meat products.Moreover,this study validated the significant impact of factors such as storage temperature,packaging methods,preservatives,and secondary sterilization on the shelf⁃life of marinated meat products from another perspective.

关 键 词：酱卤肉制品 机器学习 文本挖掘 货架期预测模型 食品微生物 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TS251.61[自动化与计算机技术—控制科学与工程]