USE OF MULTIVARIATE PREDICTION MODELS TO UNDERSTAND THE CONSUMER’S BEHAVIOR BASED ON PHYSICAL-CHEMICAL AND PHYSICAL ANALYSIS: A CASE STUDY WITH STRAWBERRY

Strawberries' quality greatly varies, and assessing their sensory quality is a costly task. Thus, this work employs physical and physical-chemical data (height, diameter, firmness, color, pH, acidity, and soluble solids) to predict the sensory quality of strawberries. Six sessions, with 30 strawberry samples in total, were performed. A total of 715 consumers evaluated the samples according to their acceptance, expectation, satisfiability, and pay intention. Two Multiple Linear Regressions (MLR) were used to predict the acceptance and the expectation based on the physical and physical-chemical measurements, while Multivariate Classification Models (LDA) were used to classify of the satisfiability (satisfied or not satisfied) and the pay intention (willingness to pay more or not) based on the acceptance scores. The MLR models to predict the consumer’s acceptance and expectation were subject to cross-validation, external validation, and y-randomization tests. All regression models achieved R² coefficients higher than .70 for both the cross-validation and external validation tests. Furthermore, they had small average errors and desired y-randomization results. The LDA models to predict the satisfaction and the pay intention had favorable results, whereas the satisfaction predictor model achieved 95% success rate on the cross-validation test and 87% on the external-validation test, while the pay intention model achieved a success rate of 73% and 63% on both tests, respectively. The remaining models achieved a 100% success rate for the cross-validation test and higher than 88% for the external validation. Our results indicate that multivariate models achieve good results when predicting the consumer’s response by easy-to-obtain physical-chemical data. Furthermore, it was possible to classify the strawberry samples regarding consumer satisfaction and pay intention using the acceptance scores. Those models can be used in the quality control of strawberries, aiding on establishing quality standards according to the consumer responses and ensuring that the strawberry commercialization considers sensory acceptance.