Modeling and Optimization of Solvent‐Based Ultrasonic‐Assisted Extraction of Glutinous Pigmented Rice Bran′s Total Anthocyanin Content Using Response Surface Methodology (RSM) and Artificial Neural Networks (ANNs)

Modeling nonlinear interactions within ultrasonic‐assisted extraction (UAE) processes for optimal anthocyanin compound recovery requires analytical approaches that extend beyond the structural constraints of polynomial approximations. In this study, a structured comparative modeling framework was developed to critically evaluate quadratic response surface methodology (RSM) and artificial neural networks (ANNs) with respect to predictive accuracy and optimization performance toward maximum response yield. A three‐level, three‐factor face‐centered central composite design (FCCD) was applied to optimize solvent‐based UAE for maximizing the total anthocyanin content (TAC) from Thai glutinous pigmented rice bran. The effects of UAE time ( X 1 : 25–35 min), UAE temperature ( X 2 : 30–50°C), and sample‐to‐solvent ratio ( X 3 : 1:15–1:25 g/mL) were systematically investigated as key UAE factors. The experimental UAE data were well fitted to a second‐order polynomial model, exhibiting a high coefficient of determination ( R 2 = 0.9742). Response surface analysis revealed that the optimal UAE conditions for achieving the maximum TAC of 4.14 mg C3G/g DW were a UAE time of 35 min, a UAE temperature of 42.73°C, and a sample‐to‐solvent ratio of 1:25 g/mL using 60 % ( v / v ) aqueous methanol containing 0.1 % citric acid. The ANN model identified the optimal UAE conditions for obtaining the maximum TAC of 4.23 mg C3G/g DW as an UAE time of 35 min, an UAE temperature of 42.31°C, and a sample‐to‐solvent ratio of 1:25 g/mL. Analysis using ANN predicted TAC response showed a higher R 2 value of 0.9797 and average absolute deviation (AAD) values of 2.78 %, compared with 3.61 % for RSM. Moreover, the root mean squared error (RMSE) for ANN was 0.042, which was lower than that of RSM at 0.047, further demonstrating the superior predictive performance of ANN. This work moves beyond conventional optimization by elucidating the structural limitations of quadratic response surface models and highlighting the adaptive learning capacity of ANN for nonlinear extraction dynamics. By establishing a comparative modeling framework, this study advances machine learning–driven optimization strategies and provides a scalable predictive approach for anthocyanin recovery, supporting the development of high‐value extracts for applications in functional foods, pharmaceuticals, and cosmetics.