Razvan Andonie, Anne Johansen,* Amy Mumma, Holly Pinkart, 
and Szilard Vajda  
*Chemistry Department, Central Washington University, MS 7539,
400 E. University Way, Ellensburg, WA 98926 (anne.johansen@cwu.edu)

Trustworthy wine quality assessment generally relies on the sight, nose, and palate of individual wine tasters of high reputation who have limited availability and capacity. Thus, finding a cost-efficient method to assess wine quality without relying on a human expert taster provides financial and logistical benefits. We used our own database of biochemical and organoleptic analysis of 60 different Washington Cabernet Sauvignon wines to investigate using artificial intelligence to effectively assess the overall quality of the wines based on laboratory analytical techniques. Three bottles of each wine were tested. The computer program was first trained by pairing 32 wine biochemical and other characteristics with the overall wine quality assessment of an expert taster, then optimized to pick the most cost-efficient subset of the 32 parameters that could satisfactorily assess overall wine quality. Significant prediction accuracy, in particular of very high- and low-end wines, was achieved with five measured/available characteristics. In order of importance, these are: co-pigmentation, age, region, red color, and total SO2. To the best of our knowledge, this is the first study of this kind on Washington wines.

Funding Support: Central Washington University