AI Case Study

University of Melbourne researchers predict consumer beer preference with 82% accuracy using biometric measures

University of Melbourne researchers have developed a model which uses visual cues from beer consumers to predict their beer preferences. This is based on the non-intrusive biometric responses to the foam height of a freshly poured beer - foam height is correlated with higher quality beers. The research demonstrates an 82% success rate of prediction.

Industry

Public And Social Sector

Education And Academia

Project Overview

"In this study, non-invasive biometrics along with a sensory questionnaire were used to assess consumers perception to visual attributes of 15 beers from pouring videos obtained using the RoboBEER (automatic robotic pourer). The sensory session was conducted with 30 participants using an integrated camera system, which consists of an infrared thermal camera and video recording coupled with a Bio-sensory computer application and an eye tracking device. Objective physical parameters from beer pouring were obtained using the RoboBEER and computer vision algorithms. Finally, a liking model using machine learning techniques was developed based on biometrics from consumers in combination with auto-mated outputs from the RoboBEER, which rendered high accuracy in the classiﬁcation of beers between low and high liking."

Reported Results

"The highly accurate (82%) ANN model obtained to visually assess the level of liking of foam height (low and high) by using only the subconscious responses (biometrics) and objective physical parameters as inputs could potentially be a rapid tool to assess consumers acceptability of the beer samples. This would aid in the visual screening of samples for new products development to reduce the number of potential options, and to decrease the number of descriptors to assess in the sensory questionnaires."

Technology

Computer vision algorithms were used to identify 13 parameters from beer and foam dynamics; participant heart rate assessment determined via video using Eulerian Magniﬁcation algorithm.
"Different machine learning algorithms (three decision trees, two discriminant analyses, one logistic regression classiﬁer, six support vector machines, six nearest neighbor classiﬁers, and ﬁve ensemble classiﬁers) were tested for pattern recognition, in which artiﬁcial neural networks (ANN) was assessed as the most accurate to classify the beer samples according to consumers liking."

Function

R And D

Core Research And Development

Background

"Visual descriptors are among the most important quality traits for consumers when assessing or selecting a food or beverage as they create the ﬁrst impression of the product. Usually, consumers assess a product, ﬁrst by its label and packaging, followed by the appearance of the food or beverage before they decide whether to taste it or not. In contrast to other beverages, the visual characteristics of beer that determine its quality and that are more appealing to consumers are foamability, which is the capacity of beer to form foam, foam stability, and clarity. Other sensory attributes such as ﬂavor and taste, aromas, and mouthfeel that are important for consumers, are also inﬂuenced by the foam-related parameters."

Benefits

Data

Eye-tracking and infrared camera image data (body temperature) from 30 participants taken every 2 seconds, RGB images.
"The videos for the 15 beer samples used for the sensory session obtained from the robotic pourer RoboBEER were analyzed in triplicate (three bottles of each sample) for foam and color-related parameters."