Artificial Neural Network and response surface methodology modeling-based optimization approach for the ultrasonication-assisted extraction of phenolics from mustard (BARI Sharisa-18) seed meal

• Mustard variety BARI Sharisha-18 used for efficient antioxidant extraction. • UAE optimized for extracting TPC, TFC, and DPPH radical scavenging activity. • Three optimization approaches (RSM-SA-DF, RSM-GA-DF, ANN-GA-DF) were used. • Model predictions closely matched experimental results, proving reliability. • UAE outperformed conventional extraction, with higher yields and efficiency. This study utilized a mustard variety known as BARI Sharisha-18 to efficiently extract antioxidants like TPC, TFC, and DPPH radical scavenging activity using ultrasound-assisted extraction (UAE) by optimizing key parameters, such as ultrasound temperature (U-temp), sonication time (S-time), solvent-to-material ratio (S/M), and ethanol concentration (EtOH). Response Surface Methodology (RSM) and an Artificial Neural Network (ANN) multilayer perceptron model were employed to improve reliability, with hyperparameter tuning focused on the number of neurons and activation functions to prevent overfitting. A Box-Behnken design was applied to train both models, varying the parameters: U-temp (40–70°C), S-time (20–60 minutes), EtOH (40–80%), and S/M (10–30 mL/g). Multi-objective optimization for TPC, TFC, and DPPH was managed using a desirability function (DF), leading to three optimization approaches: RSM-SA-DF, RSM-GA-DF, and ANN-GA-DF. RSM-SA-DF and RSM-GA-DF produced similar optimal conditions: S/M of 20.4 ml/g, EtOH concentration of 80%, U-temp of 40°C, and S-time of 60 minutes, yielding TFC of 225.2 mg, TPC of 1677 mg, and DPPH activity of 82%. ANN-GA-DF showed slight variations with EtOH at 67% and S/M at 15 mL/g, while maintaining U-temp at 40°C and S-time at 60 minutes, achieving TFC of 270.4 mg QE/100 g, TPC of 2092.4 mg GAE/100 g, and DPPH activity of 89%. In comparison, conventional solvent extraction (60°C for 120 min with 80% ethanol at 30 mL/g) yielded 1278.17 mg GAE/100g of TPC, 135.97 mg QE/100g of TFC, and 68.29% DPPH activity, significantly ( p > 0.05) lower than the yields achieved with UAE. Overall, this study demonstrated the strong predictive accuracy of these hybrid modeling-based optimization approaches and affirmed UAE as a robust, efficient, and sustainable alternative to CSE by yielding higher antioxidant content under milder conditions.