Prediction Model for Eating Property of Indica Rice

Abstract One hundred and five indica rice samples were analyzed by determining head rice yield, chalkiness, protein content, gel consistency and amylose content. Taste scores of rice were obtained by sensory evaluation. Support vector machine ( SVM ) and K‐nearest neighbors ( KNN ) models were established with physicochemical properties as attributes. In each linear correlation, amylose content had highest coefficient with taste score, but in the mutual influence, gel consistency showed highest correlation. In SVM model, the accuracy of the training set and testing set were 99.0% and 93.3%, respectively; in KNN model, they were 74.3% and 73.3%, respectively. These results showed that the nonlinear relationship between eating property and physicochemical properties of indica rice was found. It was concluded that SVM , which extracts nonlinear features, can be used to effectively predict the taste class of unknown. Practical Applications Support vector machine ( SVM ) is an effective prediction model used to represent the eating property of indica rice. The model showed that the nonlinear relationship between eating property and physicochemical properties of indica rice. The results of the present study can also serve as a valuable resource to further explore the utilization of the prediction model in sensory evaluation of rice, and expound the relationship between eating property and other properties of rice.