Correlation between response surface methodology and artificial neural network in the prediction of bioactive compounds of unripe Musa acuminata peel

Abstract

The objective of this research is to study the correlation between response surface methodology (RSM) and artificial neural network (ANN) on the antioxidant activity and phenolic compound extracted from unripe Musa acuminata peel waste by microwave-assisted extraction (MAE). Distilled water was used as a substitute for organic solvents throughout the extraction process to obtain toxic-free polar bioactive compounds. DPPH free radical assay and Folin–Ciocalteu methods were used to measure the antioxidant activity and the total phenolic compounds (TPC), respectively. The optimal conditions were obtained at 650 W of microwave power, 110 s of extraction time, and 0.06 g/ml of solid-to-solvent ratio. The optimized MAE extraction yielded 76.81% and 418.46 mg GA/100 g extract of antioxidants and TPC, respectively. Overall, the solid-to-solvent ratio is the most significant variable, followed by the extraction time. Both the RSM and ANN models have delivered good quality prediction for antioxidants and TPC. Nonetheless, the ANN model showed higher predictive potential due to its higher coefficient of determination (R2ANNantioxidant = 0.9803, R2RSMantioxidant = 0.9742), lower root-mean-square error (RMSEANNantioxidant = 2.69, RMSERSMantioxidant = 3.07), and lower absolute average deviation (AADANNantioxidant = 3.38%, AADRSMantioxidant = 5.64%) for both the antioxidants and the TPC. The incorporation of ANN into the RSM model in this study has successfully countered the drawbacks of RSM.

Keywords

• Antioxidants
• Microwave-assisted extraction
• Optimization
• Peel
• Phenolics