Clustering in wineinformatics with attribute selection to increase uniqueness of clusters

Abstract

Wineinformatics is a new data science research area that focuses on large amounts of wine-related data. Most of the current Wineinformatics researches are focused on supervised learning to predict the wine quality, price, region and weather. In this research, unsupervised learning using K-means clustering with optimal K search and filtration process is studied on a Bordeaux-region specific dataset to form clusters and find representative wines in each cluster. 14,349 wines representing the 21st century Bordeaux dataset are clustered into 43 and 13 clusters with detailed analysis on the number of wines, dominant wine characteristics, average wine grades, and representative wines in each cluster. Similar research results are also generated and presented on 435 elite wines (wines that scored 95 points and above on a 100 points scale). The information generated from this research can be beneficial to wine vendors to make a selection given the limited number of wines they can realistically offer, to connoisseurs to study wines in a target region/vintage/price with a representative short list, and to wine consumers to get recommendations. Many possible researches can adopt the same process to analyze and find representative wines in different wine making regions/countries, vintages, or pivot points. This paper opens up a new door for Wineinformatics in unsupervised learning researches.