Application of classification models in screening superior rice grain quality in male sterile and pollen parents

Abstract

Rice breeders often produce high-yielding hybrids targeting the needs of the farmers which often leave behind cooking and eating quality (CEQ) due to lack of high-throughput phenotyping tools. To satisfy the needs of both ends, a quick classification tool was developed for identifying high-yielding hybrids with good CEQ classes through random forest, artificial neural network and support vector machine models with accuracies of 63.9 %, 69.4 % and 69.4 %, respectively. All models were created using routine grain quality parameters, pasting, texture, and starch structure properties. Among these properties, it was found out that gelatinization temperature, breakdown viscosity and pasting viscosity are the most important parameters in classification. The model was validated and found to be applicable for seeds planted in different years and it is most suitable for predicting the similar CEQ ideotypes matching male sterile and pollen parents. It was further used to identify hybrid lines which match the properties of superior quality variety, to ensure the preferences of rice consumers.