AI prediction of consumer liking and expert quality assessment of coffee from sensory data
(Room 25 C)
10 Apr 26
9:00 AM
-
9:45 AM
We explored the use of AI tools to predict consumer liking and expert quality ratings of coffee from just-about-right (JAR) and check-all-that-apply (CATA) sensory data.
We first used a robust data analysis framework to deconstruct consumer preference using a dataset where 118 consumers rated their liking of 27 black drip coffee samples. We integrated four feature-ranking methods to identify key sensory drivers, which informed the development of predictive models to forecast consumer liking. JAR acidity, JAR flavor intensity, and CATA sweetness were found to be primary drivers of liking across the population.
We then applied the same AI tools to the prediction of quality ratings using a dataset where 53 experts rated the quality of 12 specialty and commercial coffees using Coffee Cuality™. The top 10 predictors/correlators of quality were JAR flavor, acidity, roast level and color, and CATA stale/rancid (-), astringent (-), bitter (-), sweet (+), balanced (+) and burnt (-). Random Forrest was the best performing model for the prediction of quality from sensory data with a predicted vs. true quality R2 of 0.66.
The proposed analytical pipeline enables both the prediction of consumer liking and expert quality assessment of coffee from sensory (JAR and CATA) data.