What question did this study set out to answer?

The aim is to enhance the milling and physicochemical quality of extra-long rice through optimized drying conditions using machine learning.

February 5, 2026Open Access

Optimizing tempering drying conditions for milling and physicochemical quality of extra-long rice: A machine learning approach

Key Points

The aim is to enhance the milling and physicochemical quality of extra-long rice through optimized drying conditions using machine learning.
Utilized five machine learning models and Response Surface Methodology for optimization.
Considered independent variables: drying temperature, tempering time, and initial moisture content.
Simulated drying kinetics with the Weibull distribution model.
Conducted statistical analysis for response variables including head rice, amylose content, alkali spreading value, and gel consistency.
Identified optimal drying conditions: 53 °C temperature, 60 min tempering time, and 17.5 % initial moisture content.
Achieved actual values: 33 % head rice, 18.54 % amylose content, grade 6.88 alkali spreading value, and 97.40 mm gel consistency.
k-NN model had the highest predictive accuracy among machine learning models.
RSM showed good predictions, performing better than most ML models except k-NN.

Abstract

Extra-long rice is widely exported and imported. As far as quality is concerned, the milling and physicochemical quality have industrial significance. Previous studies showed that drying has a critical impact on these qualities. Therefore, this study aims to optimize the tempering drying conditions for milling and the physicochemical quality of extra-long rice. The five machine learning (ML) models and Response Surface Methodology (RSM) were employed for optimization. The drying temperature (35, 45, and 55 °C), tempering time (60, 120, and 180 min), and initial moisture content (17, 17.5, and 18 %) were all considered independent variables. The response variables were head rice (HR), amylose content (AC), alkali spreading value (ASV), and gel consistency (GC). The kinetics of the drying process were simulated using the Weibull distribution model. The effective diffusion coefficient () values ranged from 1.04 × 10−10 to 2.73 × 10−10. The R2 and RMSE ranged from 0.81 to 0.98 and 0.04 to 0.18, respectively, suggesting good Weibull model fitting. Statistical analysis showed that the HR, AC, and ASV were highly significant (p < 0.01) at the quadratic level, whereas GC was significant (p < 0.05). Among all ML models, k-NN provided the most accurate predictions, with R2 values of 0.9105, 0.9898, 0.9989, and 0.8694 and corresponding MAE values of 1.0909, 0.2554, 0.0317, and 2.8894 for HR, AC, ASV, and GC, respectively. The RMSE values for these parameters were 1.3118, 0.2721, 0.0382, and 3.6571. The optimal drying conditions were determined to be a drying temperature of 53 °C, a tempering time of 60 min, and an initial moisture content of 17.5 %. At these conditions, the actual values for HR, AC, ASV, and GC were 33 %, 18.54 %, grade 6.88, and 97.40 mm, while the k-NN model-predicted values were 33 %, 18.83 %, grade 6.88, and 95 mm, respectively. The RSM also showed better prediction than all ML models except for the k-NN model. It was found that the k-NN and RSM models proved to be highly effective in predicting optimal tempering drying conditions for rice. This study provides valuable insights into technological advancements and industrial applications in rice processing.

Read Full Paperexternally

Mark Helpful

Bookmark

Relay

View Full Paper