Leveraging Machine Learning to Explore and Improve Beer Flavor

The perception and appreciation of beer flavor depends on many interacting chemical compounds and external factors, and therefore proves challenging to understand and predict. As a result, the brewing industry relies on humans to evaluate flavor, which is characterized by trade-offs between consistency and costs. We believe modern machine learning can support or complement humans at this task.

We combine extensive chemical  and sensory analyses of hundreds beers to train machine learning models that predict flavor and consumer appreciation. These machine learning models significantly outperform conventional statistics and accurately predict beer features and consumer appreciation from chemical profiles, and reveal key flavor-driving compounds, including those that enhance the appreciation in beer, or the body and alcoholic character in non-alcoholic beers. When we added these compounds to both alcoholic and non-alcoholic beers, we observed notable improvements in flavor attributes, demonstrating machine learning’s value for product optimization and deepening our understanding of flavor.

In this talk, we will outline our model selection and pipeline-building strategies, discuss common pitfalls, and explore further applications of machine learning for breweries and researchers seeking to elevate beer quality and drive innovation.