Optimization of total flavonoid compound extraction from Camellia sinensis using the artificial neural network and response surface methodology

The aim of this paper was to model and optimize the process of total flavonoid extraction from green tea using the artificial neural network and response surface methodology, as well as the comparison of these optimization techniques. The extraction time, ethanol concentration and solid-to-liquid ratio were identified as the independent variables, while the yield of total flavonoid was selected as the dependent variable. Central composite design (CCD), using a second-order polynomial model and multilayer perceptron (MLP) were used for fitting the obtained experimental data. The values of root mean square error, cross-validated correlation coefficient and normal correlation coefficient for, both models indicate that the artificial neural network is better in-prediction of total flavonoid yield than CCD. The optimal conditions using the desirability function at CCD model was achieved for the extraction time of 32.5 min, ethanol concentration of 100% (v/v) and solid-to-liquid ratio of 1:32.5 (m/v). The predicted yield at these conditions was 2.11 g/100 g of the dried extract (d.e.), while the experimentally obtained yield was 2.39 g/100 g d.e. The extraction process was optimized by the use of the simplex method for the MLP model. The optimal value of total flavonoid yield (2.80 g/100 g d.e.) was achieved after the, extraction time of 27.2 min using ethanol concentration of 100% (v/v) at solid-to-liquid ratio of 1:20.7 (m/v). The predicted response value under optimal conditions for MLP model was also experimentally confirmed (2.71 g/100 g d.e.).