Rheological modeling of marjoram fortified rice dough: Empirical and machine learning approach

摘要

The rheological properties (viscosity) of marjoram fortified rice dough (MFRD) insideextruder are helpful for product development, process control, and product quality.This study aims to predict viscosity of MFRD through regression analysis of existingempirical models and machine learning (ML) techniques. At first, MFRD was subjectedto steady shear analysis (SSA) at various temperatures (60–100℃) and shear rates(1–50 s~（-1）). The SSA data was split into training and validation sets in 70:30 ratio.Rheological modeling was conducted using various empirical models. Hyperparametertuning was performed using existing MATLAB functions to develop ML models. TheSSA revealed that MFRD exhibited pseudoplastic behavior, with viscosity decreasingwith increasing temperature up to 70 C and then increasing again, most probably dueto starch gelatinization. The regression analysis indicated satisfactory results forPower Law, Carreau, Cross, Sisko, Carreau-Yasuda model (R2: 0.9849–0.9992) wasdeemed the best based on Akaike information criterion. A dual variable model (DVM)was then developed using this model, and its coefficient was calculated more efficientlythrough prepareSurfaceData (R~2: 0.8865) MATLAB function than lsqcurvefit(R~2: 0.8280) function. In ML model development, Levenberg–Marquardt was themost effective for artificial neural networks (ANNs), grid search for support vector,and Bayesian optimization for decision tree, ensemble, and gaussian process regression(GPR). Only ANN and GPR models had R2(testing) of 1 during training, but GPRmodel (Nash-Sutcliffe efficiency: 0.9995) outperformed others during validation.Therefore, GPR model can be used to accurately predict MFRD viscosity and DVMfor simulation study.