ESTIMATING SENSORY TEXTURE OF COOKED RICE USING FULL AND OPTIMIZED PREDICTIVE REGRESSION MODELS

Youngseung Lee; Han Sub Kwak; Marura Lenjo; Jean François Meullenet

doi:10.9755/ejfa.2015-09-793

Authors

Youngseung Lee Department of Food Science and Nutrition, Dankook University, Yongin-si 448–701, Korea;
Han Sub Kwak Department of Food Science and Nutrition, Dankook University, Yongin-si 448–701, Korea;
Marura Lenjo Department of Food Science, University of Arkansas, Fayetteville, AR 72701 USA; †These authors contributed equally to this study
Jean François Meullenet Department of Food Science, University of Arkansas, Fayetteville, AR 72701 USA; †These authors contributed equally to this study

DOI:

https://doi.org/10.9755/ejfa.2015-09-793

Keywords:

Rice, Texture, Estimation, Partial least square regression, Jackknife resampling

Abstract

Sensory texture characteristics of cooked rice were predicted with a texture analyzer using a full predictive model (partial least square regression; PLSR) and an optimized predictive model (jackknife resampling method; JRM). Texture parameters of 102 cooked rice samples were measured using a spectral stress strain analysis. Eleven sensory texture characteristics were evaluated using a trained descriptive panel. JRM showed slightly better prediction for sensory texture attributes than PLSR due to the removal of insignificant variables. The following four sensory attributes were strongly predicted by JRM based on the calibration model correlation coefficient (Rcal): cohesion of bolus (Rcal = 0.78), adhesion to lips (Rcal = 0.83), cohesiveness (Rcal = 0.69), and hardness (Rcal = 0.72). Cohesiveness, toothpull and toothpack were moderately predicted (Rcal ≥ 0.60). The results from the texture analyzer were able to estimate sensory texture attributes, which were directly related to texture characteristics such as hardness, stickiness, cohesiveness, etc.