A 1978 Chateau Margaux retails for a mean of $541 nowadays. Among the younger wines, is it possible to tell which ones will age gracefully and become the next Chateau Margaux? TADA can help wine buyers identify the future champions.

Industry

Wine

Project Duration and Effort

Two days

Type of Prediction

Multiclass classification

Customer Benefits

  1. 91% accuracy in predicting which wines will improve with age.
  2. Identification of the chemical components helping a young wine become a great wine when aging.
  3. Speed, once the tool is in place, TADA’s analysis and new prediction takes a few minutes.

Problem to solve

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 $21.9 million, exceeding the high presale assessment by 46 percent. And amongst the star wines sold were ten bottles of Château Mouton-Rothschild 1945. They went for $343,000, hence doubling the presale estimate of $120,000.

With a dataset comprising the physicochemical information about different types of wines, we asked TADA to predict the future wine quality by sorting them into three categories: good, bad, and average.

The physicochemical criteria are the following:

  • Fixed acidity,
  • Volatile acidity,
  • Critric acid,
  • Residual sugar,
  • Chlorides,
  • Free sulfur dioxide,
  • Total sulfur dioxide,
  • Density,
  • PH,
  • Dulfates,
  • Alcohol.

Objectives

  • Predict which wines will become good, bad, or average with age with a reasonable accuracy.
  • Identify in the young wines which criteria influence wine quality.
  • Give the means to simulate in real-time the impact of each of these criteria.


This challenge raises the following wine question:

Can we predict which young wines are going to become stars?

Solution

TADA managed to identify which wines were going to become good, which wines were going to become bad, and which wines would become average with an accuracy of 91%, a MCC of 85%, and a Cohen’s Kappa coefficient of 85%.

TADA gives the means to identify, among young wines, which one will be the next Château Mouton-Rothschild 1945.

TADA has selected seven main Physico-chemical criteria out of the eleven available in the dataset to make its decision, with variable global influence on the final classification. They are the following:

  • Alcohol, with a weight in TADA’s decision of 17%,
  • Volatile acidity, with a weight in TADA’s decision of 16%,
  • Density, with a weight in TADA’s decision of 15%,
  • Chlorides, with a weight in TADA’s decision of 14%,
  • Total sulfur dioxide, with a weight in TADA’s decision of 14%,
  • Sulfates, with a weight in TADA’s decision of 13%,
  • free sulfur dioxide, with a weight in TADA’s decision of 9%.

Live Predict

It is interesting to note that the seven Physico-chemical criteria selected have approximately the same weight in the decision, meaning that it is a suitable combination of these criteria, resulting in a good wine.

Customer Benefits

In two days, wine buyers obtained:

  • 91% accuracy in predicting which wines will grow better with age.
  • Identification of the chemical components helping a young wine become a great wine when aging.
  • Dynamic analysis of the impact of each factor using Live Predict.