The prediction models among instrumental data, descriptive data, and hedonic data
2020/05/27
A. Background:
There were several studies employ instrumental analytical data to predict consumer tests data in the past. Four different tea drinks were employed to collect the data for physical /chemical analyses, descriptive sensory analysis, and consumer hedonic test. The predicting powers of physical /chemical data to consumer data and descriptive data to consumer data were compared in this study.
B. Methods:
Tea drinks with 4 different processes with significant different properties were employed as the samples. The physical/chemical analyses (including ionic, pH, turbidity, polyphenols), descriptive test (including 19 sensory descriptors), and consumer hedonic test (including 6 sensory attributes) were employ for collecting data on the 4 samples.
The data were analyzed with the multivariate analysis models, PLS and L-PLS, to check the model predicting power by the R2 as the index. The Unscrambler® X was employed for the multivariate analysis.
C. Results:
1.The R2 of the model of employing physical/chemical data predicting consumer test data was R2=0.27. This indicates that the predicting power is low.
2. The R2 of the model of employing descriptive test data predicting consumer test data was R2=0.54. This indicates that the predicting power is medium.
3. The R2 of the model of employing physical/chemical data and descriptive test data together predicting consumer test data was R2=0.33. This indicates that the predicting power is low.
4. The R2 of the model of employing physical/chemical data predicting descriptive test data was R2=0.99. This indicates that the predicting power is high.
D. Conclusion:
1.This study partially prove the theory of [using physical/chemical data to predict descriptive test data and using descriptive test data to predict consumer test data] raised by international sensory scientists in the past.
2.PLS and L-PLS are powerful tools to distinguish consumers subgroups among samples.
3.This study also shows that well trained panelists are necessary for the descriptive test.
