Accurate prediction of the eating and cooking quality of rice using artificial neural networks and the texture properties of cooked rice

Accurate prediction of the eating and cooking quality (ECQ) of rice is of great importance. Statistical and machine learning models were developed to predict the overall acceptability of cooked rice. The results showed that the models developed using stepwise multiple linear regression, principal component analysis plus multiple linear regression, partial least square regression, k-nearest neighbor, random forest, and gradient boosted decision tree had determination coefficients (R2) of 0.156-0.452, 0.357, 0.160-0.460, 0.192-0.746, 0.453-0.708, and 0.469-0.880, respectively, which were improved to 0.675-0.979 by artificial neural networks (ANN) models. The ANN models also had lower root mean square errors (0.574-1.32). Further, the ANN model using textural properties could accurately predict 92.1 % of overall acceptability, which could be improved to >96 % using the components and/or pasting characteristics. Overall, the accuracy of ECQ prediction was substantially improved by the model developed using ANN with texture properties of rice.