Can wine quality be ‘predicted’ using machine learning?
Only a small percentage of the wines produced around the world increase in value. And to be sold successfully, the wines must be impeccably stored, have traceable provenance and generally be offered in their original wooden cases. To reach such a high quality, hence price level, the quality of the wine and its potential has to be evaluated when young. If specialized wine tasters have the skills to identify these high potential wines, production managers, wine makers, resellers, traders, could use some support in assessing the potential of a young wine. The choice of a ‘future’ high quality wine is very complex.
Wine production and quality control managers can get actionable predictive guidance to help them control the quality of their wine all along the production process through the use of machine learning. However, they are usually not data scientists and they often do not have the required skills in machine learning or the software coding experience to build predictive models. Most of the time, they have historical data about their past productions and/or other vineyard productions, such as the acidity of the young wine, the volatile acidity caused by bacteria, the citric acid generally present in very small quantities in wine grapes, the residual sugar, chlorides which are usually an indicator of the wine’s “saltiness”, the sulfur dioxide as the most common preservative used and all other wine components measured all along the production process including pH and alcohol.
The data used by wine professionals are Small Data, meaning they often contain historical data only about a few dozens or hundreds of different wines but rarely thousands or more like in Big Data. Traditional machine learning tools work well with Big Data but do not perform as well with Small Data.
MyDataModels allows wine professionals to easily build predictive models from Small Data. Wine professionals do not need any special training. They can use their collected data directly, without normalization, outlier’s management or feature engineering. Thanks to this limited data preparation, the results from this specific dataset were obtained with a few clicks in 3 minutes on a regular laptop.
MyDataModels brings a self-service solution for those who have Small Data and are not data scientists.
When it comes to winemaking, know how is everything. No machine can ever replace a vintner’s palate, nor the knowledge which has been passed from generation to generation within a family of wine savvy gourmet producers. Not everybody in the wine industry has the same amount of knowledge. Nor can every vineyard afford the same level of wine connoisseur’s as wine managers that the Chateau Mouton-Rothschild vineyard can afford.
Yet it is possible, accessible and efficient for wine professionals to use machine learning in order to anticipate wine quality. For instance, in this red wine quality prediction use case, the results obtained from MyDataModels’ predictive models are very good with a 0.63 Root Mean Square Error. It means that in the estimation of the wine quality grade ranging between 0 and 10, the average error made is of 0.63 point out of 10, or in other words 6.3%. An excellent result.
“Predictive model reached excellent result on wine quality prediction”
In May 2016, Sotheby’s New York auctioned 20,000 bottles of fine and rare wine. The seller was billionaire collector William Koch. The cellar went for a total of $21.9 million, surpassing the presale high estimate by 46 percent. And among the star wines sold were 10 bottles of Château Mouton-Rothschild 1945. They went for $343,000, therefore doubling the presale estimate of $120,000.
Using MyDataModels application empowers most vine professionals to produce or identify the next Château Mouton-Rothschild 1945.