Study on Wine Evaluation Based on LASSO Regression

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,pH 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.