3.1.2. Optimization of a Suitable Extraction Method
A suitable extraction method for the MAE technique using BBD was evaluated. The ranges of Y
1 and Y
2 were 0.00–20.22 mg/g and 0.00–5.62%, respectively, as shown in
Table 3. From 27 experiments, 4 experiments (F6, F11, F16, and F24) did not yield enough extract for further mangiferin content evaluation, while 2 experiments (F8 and F23) had missing data because some replications yielded extracts lower than 15 mg (only one replication could be evaluated). Moreover, F13, F26 and F27 were performed according to BBD as the baseline of four variables. Among all experiments, F15 showed the highest mangiferin content, followed by F10 and F12 at 20.22 ± 0.38, 18.57 ± 0.34, and 17.28 ± 0.95 mg/g extract, respectively. The mango peel with the highest mangiferin content from F15 (fresh 100 g equal to dry power 20 g) was extracted using a microwave power of 450 W for 3 min with 50% EtOH in water (120 mL) as the solvent for extraction.
The data collected from all 27 experiments were employed to concurrently establish the second-order polynomial equation as depicted in Equation (1). Furthermore,
Table 4 provides correlation values, coefficients of determination (R
2), adjusted coefficients of determination (adjusted R
2), and ANOVA results. These tables also include the regression equations formulated for each response variable.
In this study, the contour plots illustrated the interaction between two factors displayed as a two-dimensional graph. The contour plots for the six responses under investigation, namely, Y
1 and Y
2, can be found in
Figure 1 and
Figure 2, respectively. For Y
1, focusing on its impact on the mangiferin content, the optimization suggests a subsequent linear equation.
Based on the ANOVA results, the selection of a linear model for the mangiferin content was found to be the most appropriate choice compared to two-factor, quadratic, or cubic models. This preference is substantiated by the linear model’s F value of 2.77, indicating that the model terms are statistically significant at a
p value = 0.0423 (
p value < 0.05), suggesting a reasonable fit to the experimental results (
Table 4). The linear terms of solvent ratio (X
1), extraction power (X
2), and extraction time (X
3) exhibited a significant influence on the mangiferin content, as indicated by their p values (0.0008, 0.0374, and 0.0401, respectively). Other terms in the model do not show significant effects. The lack of fit value (F value = 0.56) was nonsignificant due to noise (
p > 0.05), ensuring the validity of the model. This suggests that the linear model captures all the variability in the data, with a 78.29% chance that such a discrepancy could be due to random noise. The N-probability plot, each point aligning along a straight line with an “S” shape, demonstrates that the residuals conform to a normal distribution (
Figure 3). This pattern suggests that employing response transformation can lead to a valid analysis. Consequently, this model is suitable for the navigation of the design space. The contour plots (
Figure 1) illustrate the influence of various independent variables on the mangiferin content (Y
1).
For Y
2, focusing on its impact on the % yield of the extract per 100 g of fresh mango peel, the optimization suggests the subsequent linear equation.
Based on the ANOVA results, the selection of a linear model for the % yield was found to be the most appropriate choice compared to two-factor, quadratic, or cubic models. This preference is substantiated by the linear model’s F value of 5.17, which, with a
p value lower than 0.05 (indicating that the model terms are statistically significant), suggests a reasonable fit to the data. Notably, only the linear terms of solvent ratio (X
1), extraction power, and X
3 exhibited a significant influence on the % yield, as indicated by their
p values (<0.0001 and 0.0059, respectively). Other terms in the model do not show significant effects. However, a notable concern is the significant lack of fit, as evidenced by an F value of 71.22 (
p value = 0.0139). This suggests that the linear model might not be capturing all the variability in the data, with a 1.50% chance that such a discrepancy could be due to random noise. This significant lack of fit indicates that while the linear model is the best among the models tested, it might not be an ideal representation of the underlying process. For future studies, it is advisable to explore other modeling approaches or refine the current linear model. This could involve investigating potential interaction effects, nonlinear relationships, or additional variables that were not included in the current model but may have a significant impact on the % yield. A more comprehensive model that reduces the lack of fit could lead to more accurate predictions and a deeper understanding of the factors influencing the yield. Additionally, further validation with different data sets would be crucial to ensure the robustness and generalizability of the model. The N-probability plot, each point aligning along a straight line with an “S” shape, demonstrates that the residuals conform to a normal distribution (
Figure 4). This pattern suggests that employing response transformation can lead to a valid analysis. Consequently, this model is suitable for the navigation of the design space. The contour plots (
Figure 2) illustrate the influence of various independent variables on % yield (Y
2).
In this study, we optimized a suitable extraction method from a total of 27 experiments. The ideal formulation was determined using a response optimizer plot with a composite desirability (D) value of 0.9843. The optimization plot, displayed in
Figure 5, reveals the influence of each parameter on the responses or composite desirability (rows). The vertical red lines on the graph indicate the current parameter settings, while the horizontal blue lines represent the responses corresponding to those parameter levels. To achieve the best results, two parameters were adjusted to their maximum levels, specifically mangiferin content and % yield. Based on these criteria, the optimal formulation comprised a solvent ratio of 120 mL per 100 g of sample, extraction power of 450 W, an extraction time of 4.335 min, and an EtOH ratio (EtOH in water) of 69.4444%. Subsequently, a small-scale extraction experiment was conducted using the same parameters: solvent ratio of 120 mL, extraction power of 450 W, extraction time of 4 min, and an EtOH ratio of 70% EtOH (n = 3).
The resulting extract showed a mangiferin content of 27.24 ± 2.05 mg/g and a fresh peel yield of 3.71 ± 0.17%. These results show the difference from the BBD analysis estimates that gave a mangiferin content of 19.62 mg/g and a yield of fresh peel of 5.61%. From the actual experiments, the data show that this method is a suitable extraction method for extracting mangiferin from fresh peels of mango due to the higher mangiferin content, even if it decreases the yield. Moreover, the extraction method on a small scale was prepared on a pilot scale by using a microwave-assisted extraction machine. The extract was prepared from a fresh peel of mango (7 kg) and extracted with 70% EtOH (8.4 L) for 4 min. The results showed a mangiferin content of 51.85 ± 0.35 mg/g and a yield of fresh peel of 4.35%. These results support the higher extraction capacity of MAE compared to conventional methods (ME and HE), resulting in a more than twofold increase in mangiferin content (
Table 2), which is consistent with our previous studies on other medicinal plants [
21,
22]. Our findings show promising data that support the use of an optimized extraction method for mango peel waste utilization at both the small and pilot scales. This method could be applied as the most suitable extraction method for the peel of
Ma-Muang Bao from agro-industrial waste products. Moreover, the extract obtained from this method was set as standardized mangiferin-rich mango peel extract (SMPE).