Construction of a machine learning-based prediction model for rice varieties—cooking—mastication

Mengjie Ma, Zhengbiao Gu, Li Cheng, Zhaofeng Li, Caiming Li, Mingyi Shen, Jinyi Wu, Jia Liu, Yan Hong,
Construction of a machine learning-based prediction model for rice varieties—cooking—mastication,
Food Research International,
Volume 222, Part 1,
2025,
117705,
ISSN 0963-9969,
https://doi.org/10.1016/j.foodres.2025.117705.
(https://www.sciencedirect.com/science/article/pii/S0963996925020435)
Abstract: This study established a machine learning-driven, multi-scale predictive framework to quantify the relationships among rice physicochemical properties, cooking conditions, and oral processing. 56 rice varieties were characterized for texture, morphology, and amylose content, while in vivo mastication parameters (chewing time, number of chews, saliva volume) and bolus properties (particle size, reducing sugar content) were measured. Correlation analysis identified hardness, gumminess, and chewiness as key determinants of mastication behavior (r > 0.7), whereas grain length-breadth ratio negatively influenced bolus particle size (r > 0.7). Five prediction models, including multiple linear regression, extreme gradient boosting (XGBoost), support vector machine (SVM), long short-term memory network (LSTM), and convolutional neural network, were developed, with LSTM achieving optimal mastication prediction (test R2 = 0.9499) and SVM excelling in bolus property forecasting (R2 = 0.9168). An XGBoost model effectively captured nonlinear interactions between cooking parameters (cooking time, water ratio) and texture (R2 = 0.8161). Cascade models (XGBoost-LSTM for mastication, XGBoost-SVM for bolus) enabled end-to-end prediction from cooking inputs to oral outcomes. This framework advances digital optimization of rice processing and sensory quality, offering a predictive tool for tailored food design.
Keywords: rice; oral mastication; texture properties; machine learning; prediction model; in vivo