Modeling and optimization of capsaicin extraction from Capsicum annuum L. using response surface methodology (RSM), artificial neural network (ANN), and Simulink simulation

The heartening applications of capsaicin (CAP) in agro-food, cosmetic, and pharmaceutical sectors impose their strengthening by an appropriate extraction protocol. Herein, response surface methodology coupled with desirability function (RSM-DF), artificial neural network coupled with genetic algorithm (ANN-GA), and Simulink models were developed to optimize CAP extraction from pepper (Capsicum annuum L.). The effect of four process variables, drying temperature, sample/solvent ratio, solvent, and extraction time were investigated using I-optimal design. The higher CAP content (0.0163 mg/g DW) was recorded with the following conditions: 90 degrees C drying temperature, 54 g/L concentration, and 48.75 min of extraction with acetonitrile. Drying temperature and sample/solvent ratio are the most influencing variables for the highest CAP recovery, whereas, no effect was noted for solvent factor. The three models have been successful in predicting the CAP content within the range of experimental variables. Nevertheless, ANN prediction is more accurate than RSM and Simulink with higher coefficient of determination (R-2) (0.9901 vs. 0.9602 and 0.9607, respectively) and lower mean squared error (MSE) (1.19E-07 vs. 3.54E-07 and 3.49E-07), root mean squared error (RMSE) (3.45E-04 vs. 5.95E-04 and 5.91E-04), and absolute average deviation (AAD) (2.142 % vs. 3.377 and 3.590 %). Superiority of ANN-GA approach was evidenced as well in the optimization step with higher recovery and less variation range between optimized and validated values.

Automated summary

This study optimized the extraction of capsaicin from peppers using response surface methodology, artificial neural network, and Simulink models, with artificial neural network proving to be the most accurate method with higher recovery and less variation range in the optimization step.