Brewing a good tasting, high-quality beer in a reproducible manner is a very complex and challenging process. A wide range of parameters and factors effect beer quality in nonlinear ways. These include: the composition of the grain bill; the selection of the hops added; the selection of yeast strain(s); the composition of the water used; and optimization of timing and temperature. Here we develop several models that input such features and generate a prediction of beer quality. For this purpose, beer recipes and associated quality rankings (e.g., 1-5 beers, with 5 being the best) were downloaded from www.BeerSmith.com and www.BrewersFriend.com and converted into descriptors. These features were fed into to Molegro Data Modeller and normalized. Feature selection helped reduce the possible input space. We evaluated several nonlinear and/or machine learning modeling methods using features from BrewersFriend for their ability to predict beer quality at BeerSmith. These included: Multiple Linear Regression; Partial Least Squares Regression; Support Vector Machines; Neural Networks; and KNN Classification. The resulting models demonstrated the ability to associate beer recipe features with quality and can be used in the production of novel beers.

Modeling beer ratings with machine learning

ACS Spring 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.