Fuzzy models to predict consumer ratings for biscuits based on digital image features

Fuzzy models to recognize consumer preferences were developed as part of an automated inspection system for biscuits. Digital images were used to estimate physical features of chocolate chip cookies including size, shape, baked dough color, and fraction of top surface area that was chocolate chips. Polls were conducted to determine consumer ratings of cookies. Four fuzzy models were developed to predict consumer ratings based on three of the features. There was substantial variation in consumer ratings in terms of individual opinions (30 panelists in each poll) as well as poll-to-poll differences (three calibration polls). Parameters for the inference system, including fuzzy values for cookie features and consumer ratings, were defined based on judgment and statistical analysis of data from the calibration polls. Two of the fuzzy models gave satisfactory estimates of average consumer ratings for two validation sets (44 cookies). One was a Mamdani inference system that was based on eight fuzzy values for consumer ratings. These were defined using rating distributions from calibration polls. The second model was a Sugeno inference system developed using the adaptive neurofuzzy inference system (ANFIS) algorithm (MatLab®Version 5.2, The MathWorks Inc., Natick, MA) with the calibration poll data.