文献类型：期刊文献

中文题名：基于LASSO回归的葡萄酒评价研究

英文题名：Study on Wine Evaluation Based on LASSO Regression

作者：姚燕云[1,2];蔡尚真[3]

机构：[1]浙江工商大学数量与统计研究所;[2]绍兴文理学院数学系;[3]绍兴文理学院元培学院

年份：2016

卷号：31

期号：2

起止页码：294

中文期刊名：云南农业大学学报：自然科学版

收录：CSTPCD、、北大核心2014、北大核心、CSCD_E2015_2016、CSCD

基金：2013年全国大学生数学建模赛题后续研究项目(A1208);绍兴文理学院教学成果奖培育项目;绍兴市高等教育教学改革项目

语种：中文

中文关键词：葡萄酒评价;LASSO回归;多重共线性;可信度检验

外文关键词：wine evaluation;LASSO regression;multicollinearity;reliability test

中文摘要：葡萄酒消费已融入大众生活,其质量及评价也将得到越来越多人的关注。现有评价以评酒员的感官评价为主,但往往较烦琐且费用高昂。文献的研究表明葡萄酒质量与其理化指标休戚相关,但理化指标众多,因此本研究选取能降维与处理"小n大p"问题的LASSO回归进行建模,挖掘葡萄酒质量的重要影响因素。本研究分别建立了基于外观、香气、口感和平衡/整体的四方面评价模型和总分评价模型,LASSO回归结果表明两模型都有较高精度,但总分评价模型更优,影响葡萄酒质量主要为八个指标:芳香物质的3-甲基-1-丁醇、辛酸乙酯、丁二酸二乙酯、癸酸乙酯、乙酸乙酯以及酿酒葡萄的总酚、苹果酸和p H值。进一步对总分评价模型设计可信度检验,发现该模型不具有显著的样本依赖性,少量训练样本便有良好的评价效果,该方法能够有效降低评价与测试成本,相关结果可为葡萄酒评价过程提供重要参考。

外文摘要：Wine consumption has integrated into many people’s daily life and its quality together with its evaluation has gained more and more attention. At the present,wine quality is usually evaluated by some qualified members’ sensory assessments,which is often complicated and costly. It is shown that the wine quality is closely related with its physicochemical indicators in many literatures. For the indicators are too numerous,LASSO regression,which can be used to dispose dimension reduction and the problem of ＂small n and large p＂,was selected to modeling and mining the important influence factors of the wine quality in this study. Evaluation model about four aspects： visual,aroma,taste and harmony and overall evaluation model were built separately,the results of LASSO indicated that both models had good accuracy,but the overall evaluation model was better,and there were eight indicators influencing the wine quality significantly,which were 1-buanol-3-methyl,octanoic acid-ethyl ester,butanedioic acid-diethyl ester,decanoic acid- ethyl ester,ethyl acetate from the aromatic hydrocarbon substances and total phenols,malic acid,p H value from grapes. Moreover,reliability test was designed to diagnose the overall model,as a result,there was not significant sample dependence,just several training samples could ensure good effect of evaluation,and evaluation method based on such model could reduce the appraisal and test costs effectively.The result from our model can provide reference for wine evaluation process.