Microencapsulation of Dragon Fruit Peel Extract by Freeze-Drying Using Hydrocolloids: Optimization by Hybrid Artificial Neural Network and Genetic Algorithm

The freeze-drying encapsulation process was used to encapsulate dragon fruit peel extract with three distinct wall materials: maltodextrin, gum arabica, and gelatin. The process was modeled using a feed forward back propagation artificial neural network with four, eleven, and four neurons in the input, hidden, and output layers. Three of the four input neurons were concentrations of wall material (g), while the fourth was ultrasonication power. The four output neurons were encapsulation efficiency, antioxidant activity, hygroscopicity, and solubility of freeze-dried encapsulation powder. The procedure was optimized using hybrid artificial neural network (ANN) and genetic algorithm (GA) approach. The optimal wall material composition for encapsulation obtained by the integrated ANN and GA was 4.461 g maltodextrin, 3.863 g gum arabic, and 3.198 g gelatin. The optimal ultrasonication power for achieving a homogenous mixture was determined to be 123 W. At the optimal condition, the predicted values for the responses encapsulation efficiency, antioxidant activity, hygroscopicity, and solubility were found to be 88.143%, 81.702%, 6.924 g/100 g, and 32.841%, respectively. Under optimal conditions, the relative deviation between the predicted model and experimental outcomes was less than 2.077%. The thermal stability of the encapsulated powder followed the first order kinetic modeling. The results showed that the sample treated at pH of 7 was more thermally stable at 80 °C than the sample treated at pH of 3.6. The half-life time was found to be 140 min and 103 min for the sample treated at pH of 7 and 3.6, respectively.


Introduction
Dragon fruit, known for its distinct look and delicate texture, belongs to the family of Cactaceae and is a vine climbing cactus of the Hylocereus genus (Hua et al., 2018). Dragon fruit used to be a backyard plant grown only for table consumption and medicinal uses. But now, there is a global growth of dragon fruit production from which growers are getting benefitted (Balendres & Bengoa, 2019). The pulp of dragon fruit has a high concentration of vitamin C as well as water-soluble fiber. The reddish color imparted to the peels of Hylocereusundatus is associated with persistent nitrogen containing pigment known as betacyanin, which is water soluble (Al-Mekhlafi et al., 2021). The peel of the fruit was a major by-product of the dragon fruit juice processing industry and was generally dumped onto the ecosystem. Recent research on dragon fruit peel was aimed at extracting polyphenols, betacyanin (Bhagya Raj & Dash, 2020b), and pectin (Tien et al., 2022).
The temperature during food processing and storage significantly impacts betacyanin stability since heating induces the breakdown of this pigment (Gengatharan et al., 2015;Herbach et al., 2006). In addition to these factors that degrade the stability of betacyanin were light exposure, pH, water activity, soluble metals, and enzymatic reactions (Rahayuningsih et al., 2021). Degradation and low stability of betacyanin can be the major issue in use as a food colorant in the food matrix. As natural dyes' potential and drawbacks, some innovation is necessary to minimize 1 3 natural dyes' disadvantages as food colorants (Rahayuningsih et al., 2019). The utilization of betacyanin in food processing can be increased by enhancing its stability, which can be accomplished by the encapsulation method. Therefore, encapsulation of polyphenols from extract can be a substantial alternative approach for boosting the stability of bioactive substances.
Encapsulation is a technique in which targeted compounds are entrapped by an encapsulant in order to form capsules or microcapsules at a micrometer or nanometer scale and eventually released at the time of requisite (Bamidele & Emmambux, 2021). The encapsulation method was extensively applied in food processing industries to encapsulate phytochemicals such as polyphenols, micronutrients, enzymes, and antioxidants by creating protective barriers against the light, oxygen, pH, moisture, heat, shear, or other extreme conditions during processing and storage. It improves the stability of the product when transiting through the gastrointestinal tract (Dima et al., 2015). Different methods were implemented to achieve the encapsulation of phytochemicals from food materials, including wet heat processing (pasting, kneading, and co-precipitation), spray drying, spray cooling/chilling, extrusion coating, fluidized bed coating, liposome entrapment, coacervation, inclusion complexation, centrifugal suspension separation, co-crystallization, liposomes, nanoparticles, freeze-drying, emulsion, and rotational suspension separation (Boger et al., 2021;Carpentier et al., 2021;Lengyel et al., 2019). Usually, the encapsulation process consists of two main steps, such as homogeneous mixing of components and drying. Ultrasound treatment can be applied for homogeneous mixing, and it is an effective, non-toxic, and environmentally friendly process that shortens the drying time (Yazicioglu et al., 2015).
The freeze-drying encapsulation technique was commonly utilized to entrap phytochemicals from agricultural products into another material when the core molecules are heat sensitive (Cheng et al., 2017;Yamashita et al., 2017). Mass transfer in freeze-drying does not occur through evaporation but rather through freezing, sublimation, and desorption for dehydrating the material. The powder is formed when the frozen feed emulsion is sublimated. (Ozkan et al., 2019). The freeze-drying technique was successfully implemented to microencapsulate bioactive compounds from fruits and vegetables with different wall materials, such as encapsulation of betanin (a natural dye of red beets) in combination with maltodextrin and xanthan gum, bioactive compounds of Malpighia emarginata DC fruit pulp extracts using gum arabica and maltodextrin (Rezende et al., 2018), anthocyanin of purple roselle with maltodextrin as carrier agent (Nafiunisa et al., 2017), betalain extract from cactus pear fruit by starch and maltodextrin as wall materials (Morales et al., 2021), anthocyanins from pomegranate peel using maltodextrin and calcium chloride (Azarpazhooh et al., 2019), red beetroot extract using carrier agents as maltodextrin and whey protein isolate (Faridnia et al., 2020), and curcumin with various wall materials (Guo et al., 2020).
Maltodextrin is a polysaccharide that is commonly used as a carrier agent for microencapsulation due to its high solubility in water (Galves et al., 2021). The solutions formed from different dextrose equivalents of maltodextrin are colorless and have low viscosity at high solid ratio among different wall materials used for encapsulation (Akhavan Mahdavi et al., 2016;Robert et al., 2010). Gum arabic is a natural polysaccharide and exudate gum of acacia that is widely utilized as an efficient wall material in various food applications (Carpentier et al., 2022). Gum arabic is preferred as a carrier agent due to emulsion formation and excellent volatile retention ability (Rajabi et al., 2015). Gelatin has good biocompatibility and biodegradability and is a polypeptide with high molecular weight derived by regulated hydrolysis of collagen from bones and skin (Elzoghby, 2013;Nezhadi et al., 2009). It has the ability to form a film and a gel with excellent mechanical properties that also serve as barriers to oxygen, carbon dioxide, and volatile compounds (Tongnuanchan et al., 2012). Xanthan gum is colorless, tasteless, odorless, and has a smooth texture. Xanthan gum has low viscosity at high shear rates and high viscosity at low shear rates. It has stability against heat, acid, and alkali conditions and is a clear weak gel-like liquid that can be used in a variety of industrial applications (Jo et al., 2018). The use of one carrier agent or wall material for encapsulation is not preferable as it does not maintain all necessary properties to retain the core material. As a result, a wall material comprising two or more carrier agent mixtures could be employed to increase the encapsulation characteristics of the compound of interest. This could be accomplished by combining carbohydrates with proteins and polysaccharides in various proportions.
The artificial neural network is considered a powerful tool for developing mathematical models with higher accuracy and flexibility for modeling processes and predicting responses (Bhagya Raj & Dash, 2020a;Huang et al., 2017). In many of the ANN modeled food processing methods, the genetic algorithm (GA) was used for optimizing the process parameters such as microwave puffing (K. K. Dash & Das, 2021), ultrasound-assisted extraction (Bhagya Raj & Dash, 2020b), convective drying (Chasiotis et al., 2020), spray drying (Przybył et al., 2018), and microwave vacuum drying (Khawas et al., 2016). Many researchers integrated ANN with GA for modeling and optimizing the encapsulation process, such as phenolic extracts from different agricultural produces (Espinosa-Sandoval et al., 2019), blueberry anthocyanin extract (Tao et al., 2017), and olive pomace phenols (Aliakbarian et al., 2018). There has only been a little research on the effect of hydrocolloids and ultrasound on the encapsulation process. Based on this, the objective of this study is to encapsulate the phytochemicals of dragon fruit peel by using the freeze-drying technique using three different wall materials such as maltodextrin, gum arabica, and gelatin. The influence of wall materials and ultrasonication power on encapsulation efficiency, antioxidant activity, hygroscopicity, and solubility of the encapsulated product was studied using a feed-forward back propagation neural network, and the process was optimized by using integrating ANN-GA approach.

Raw Materials
Dragon fruits (Hylocereus undatus) were procured from the local markets of Tezpur, Assam, and were cleaned thoroughly to free them from foreign materials. Then the dragon fruit peel was separated from the pulp of the fruit, and the peels were dried in a freeze dryer (Lyolab Freeze Lab, Lyophilization Systems Inc., USA). The dried peels were ground into powder and passed through a mesh of size 0.25 mm.

Ultrasound-Assisted Extraction
The betacyanin from freeze-dried dragon fruit peel powder was extracted with the application of ultrasound using ultrasonic homogenizer with probe (U500, Takashi, Japan). The extraction was carried out at 100 W of ultrasonication power, 60 °C temperature controlled by a circulating water bath, 60% ethanol concentration, 25:1 mL/g of solvent to solid ratio, and 20 min of ultrasonic treatment time (Bhagya Raj & Dash, 2020b). After the extraction, the extract was stored at − 18 °C for further analysis.

Microencapsulation of Dragon Fruit Peel Extracts by Freeze-Drying
Plain maltodextrin ( X 1 ), gum arabic ( X 2 ), and gelatin ( X 3 ) were used as the encapsulating agents for the freeze-drying encapsulation method. Ultrasound power ( X 4 ) was implemented for homogeneous mixing of wall material and extract. The experiments on the encapsulation process of dragon fruit peel extract by freeze-drying were designed using the central composite design that yielded 30 experiments. The process was modeled and optimized using a combined feed-forward back propagation neural network and genetic algorithm. For the experimental design, different combinations of encapsulating agents with the range from 1 to 5 g and ultrasonication power with a range from 100 to 300 W were selected as independent variables with five levels. The effect of these four independent variables on the physicochemical properties of the freeze-dried encapsulated powder, i.e., encapsulation efficiency ( Y 1 ), antioxidant activity ( Y 2 ), hygroscopicity ( Y 4 ), and solubility ( Y 4 ), was studied, and these were selected as dependent variables or responses of the model.
The three selected encapsulating agents were formed into wall material compositions for freeze-drying encapsulation according to the design of the experiments. The mixture of the encapsulating agents was dissolved in distilled water and mixed using a magnetic stirrer at 25 °C to obtain 20% total solid concentration (Mahdavee Khazaei et al., 2014). Different volumes of distilled water were added to the different combinations of encapsulating agents presented in the Supplementary Table S1. For the 15th experimental run, the encapsulating agent consisted of 4 g of maltodextrin, 4 g of gum arabic, and 2 g of gelatin which were mixed with 40 mL of water to form 20% total solids. The solution was stored at 4 °C for 24 h to attain complete hydration. The ultrasonicassisted dragon fruit peel extract was progressively added to the solution comprising wall materials in a weight ratio of 1:1 (wall material:extract (w:w)). The mixture was mixed under vigorous magnetic stirring, and the pH value was adjusted to 2.0. To ensure complete homogenization, samples were further sonicated at different ultrasonication power according to the experimental design at room temperature for 5 min. The homogenized mixture of wall agents and core was dried using a freeze drier (Lyolab Freeze Lab, Lyophilization Systems Inc., USA). The dried samples were ground into a fine powder and then packed in polyethylene bags and stored at 4 °C for further analysis.

Modeling and Optimization
Feed-forward backpropagation ANN was applied to model the encapsulation process by freeze-drying with one layer in input, hidden, and output layers, as shown in Fig. 1. The freeze-drying encapsulation was modeled using an artificial neural network with four input neurons that are independent variables and four output neurons or responses. The experimental data sets were split into three groups, each with a proportion of 70, 15, and 15 for training, testing, and validation of the model, respectively. The neurons in the hidden layer varied from 2 to 15, and the number of neurons in the hidden layer of the ANN model was selected based on the lower MSE value of ANN training. Levenberg-Marquardt algorithm was used for the training of the ANN model. The log sigmoid function, whose range was − 1 to + 1 for input neurons and 0 to + 1 for output neurons, was used as an activation function between the input and hidden layer and the hidden and output layer. The independent neurons were the amount of maltodextrin, gum arabica, gelatin, and ultrasonication power. The real values of input neurons were designated as X 1 , X 2 , X 3 , and X 4 , respectively, and their coded values as x 1 , x 2 , x 3 , and x 4 , respectively. Similarly, the real values of dependent variables encapsulation efficiency, antioxidant activity, hygroscopicity, and solubility of the freeze-dried encapsulated product were assigned as Y 1 , Y 2 , Y 3 , and Y 4 , respectively, and their respective coded values as y 1 , y 2 , y 3 , and y 4 , respectively.
The weights between the input and hidden layer and the weights between the hidden and output layer were assigned as w hi and w oh , respectively. The bias matrix at hidden was designated as b h1 and at output layer was represented as b o1 . The initial values of weights and bias were created by the random number generation method. Forward and backward calculations were executed repeatedly on all data pairs obtained at input and output layers until lower MSE values were achieved, and the final parameters of the ANN model were used to predict the selected responses by using Eq. (1) The relative influence of each process parameter on the response was calculated as per the method described in osmotic dehydration of banana (K. Dash et al., 2020) and microwave vacuum drying (Khawas et al., 2016).
The output parameters of the proposed ANN model were fed into the genetic algorithm as the initial population for optimizing the process, illustrated in Fig. 2. The methodology of GA works on the fundamentals of biological evolution, i.e., on the "survival of the fittest" and "natural selection" strategy. The initial population was used in the proposed fitness function (Eq. 2), and the optimized condition will be selected by the subsequent selection, crossover, and mutation, as shown in Fig. 2. (1)

Encapsulation Efficiency
The encapsulation efficiency is defined as the ratio of betacyanin content inside the encapsulate to the betacyanin content present in the UAE extract (prior to encapsulation). The betacyanin content was determined using McIlvaine buffer by following the method described for ultrasound-assisted extract from dragon fruit peel (Bhagya Raj & Dash, 2020b). Briefly, the betacyanin content was determined by McIlvaine buffer, which was prepared by adding 30 mL of 0.1 M citric acid with 70 mL of 0.2 M sodium phosphate dibasic. A volume of 0.1 mL extract was added to 3.9 mL of prepared buffer. The absorbance of the mixture was measured using a UV-vis spectrophotometer (Agilent cary100) at the wavelength of 537 nm. The encapsulation efficiency was calculated by using Eq. (3) (Essifi et al., 2021).
where Be final is the absorbance of the sample after dissolution; Be initial is the initial absorbance of the sample.

Antioxidant Activity
The antioxidative activity of freeze-dried encapsulated dragon fruit peel was determined using DPPH assay (Šturm et al., 2019). Concisely, the encapsulated powder was added to the distilled water at the concentration of 0.5 mg/mL, and 0.1 mL of this solution was mixed with 2.9 mL of 0.1 nM solution of DPPH solution in ethanol of 96% purity. The mixture was kept at incubation for 30 min, followed by measurement of absorbance at 517 nm using a UV-vis spectrophotometer (Agilent cary100). The blank was prepared with distilled water instead of the encapsulated solution, and the antioxidant activity was calculated using Eq. (4).
A control = mixture of ethanol and DPPH solution, A sample = solution containing encapsulation powder and DPPH solution.

Hygroscopicity
The hygroscopicity of samples was determined by placing around 0.2 g of freeze-dried encapsulate powder in a Petri dish and storing it for 1 week in a desiccator containing saturated NaCl (RH 75%). After 1 week, the samples were weighed, and hygroscopicity was calculated as grams of water absorbed per 100 g of sample (%) (Yamashita et al., 2017).

Solubility
The solubility of encapsulated powder was determined according to the process followed by Wang et al. (2020). Briefly, about 0.5 g weight of encapsulated powder was added to 50 mL of distilled water present in a conical flask. The conical flask was kept in an orbital shaker at 100 rpm for 30 min at 25 °C to homogenize the mixture. After 30 min, the mixture was centrifuged at 3500 rpm for 5 min, and the supernatant was weighed ( w initial ) and dried in a hot air oven at 105 °C for 5 h. The dried supernatant was weighed after drying and used in Eq. (5) to calculate solubility.

Thermal Stability of Freeze-Dried Encapsulated Dragon Fruit Peel Powders
The thermal stability of the encapsulated product was determined at two pH concentrations of 3.6 (citric acid buffer) and 7 (phosphate buffer) at a temperature of 80 °C (Tao et al., 2017). The encapsulated powder (0.5 g) was dissolved in 50 mL of 0.1 mol/L citric acid buffer for pH 3.6 and phosphate buffer solution for pH 7. Any changes in the pH of the solution were adjusted after the addition of encapsulated powder. The solution was maintained at 80 °C in a water bath after the adjustments. The total betacyanin content of the solution was measured after 10, 20, 30, 45, 60, and 90 min. The data were fitted to a first-order reaction using Eq. (6), and the betacyanin half-life period was estimated using Eq. (7) .
where t is time (min), C 0 is the initial betacyanin content (mg/g), C t is the betacyanin content at time t (mg/g), k is the first-order kinetic constant ( min −1 ), and t 1∕2 is half life time (min).

Statistical Analysis
The statistical parameters were analyzed to validate the adequacy of the model. The coefficient of determination ( R 2 ) interprets the relative measure of fit and ranges from 0 to 1; a higher value of R 2 (close to unity) implies that the model is capable of predicting the responses. Chi-square ( 2 ) and root mean square error (RMSE) represent the absolute fit of the model to the experimental data. 2 is a measure of the change between the experimental and predicted data set. RMSE can be elucidated as the standard deviation of the unknown variance. Lower values of 2 and RMSE signify the better fit of the model to the experimental values. The relative deviation percentage ( R d ) validated the adequacy of the ANN model using Eq. (8) (Bhagya Raj & Dash, 2021).
where Y p is the predicted value from the model, Y e is the observed value or experimental values, and n is the number of experiments conducted.

Artificial Neural Network Modeling of Encapsulation Process
The artificial neural network was implemented to simulate the encapsulation process and predict the responses based on various combinations of independent variables. The neurons in the hidden layer were chosen based on the lower MSE value obtained for training data, and the optimal ANN topology was found to be 4-11-4 with 11 neurons in the hidden layer. The selected ANN topology resulted in a R 2 value higher than 0.989 and MSE value lower than 0.01 for training, testing, and validation data after completion of 5000 cycles. The matrix size for the weights between input and hidden layer ( w hi ) was 11 × 4 and for the weights between hidden and output layer ( w oh ) was 4 × 11. The final bias matrix at hidden ( b h1 ) and output layer ( b o1 ) was a column matrix with sizes 11 × 1 and 4 × 1, respectively.
The output at the hidden layer and output layer was represented as o h and y o , respectively, where subscript h and o represented the neuron number for the corresponding layers. The output of the output layer can be calculated using the hidden layer output, weights between the hidden and output layer, and bias at the output layer. The equation for output at the hidden layer is presented in Eq. (9).
The output for the first neuron y 1 (encapsulation efficiency) was calculated by applying log sigmoid function to the value obtained after multiplying the first row of w oh , i.e., w 1h having matrix size of 1 × 11 with the output of hidden layer ( o h ) having matrix size of 11 × 1, then added to the first row of bias matrix b o1 at the output layer, i.e., b 11 ; the output of the first neuron of the output layer is presented in Table 1.
Similarly, the output for the second, third, and fourth neurons of the output layer was calculated using the corresponding row of w oh and b o1 , as shown in Table 1.
The output for each of the hidden layer neurons (Eq. 10) was calculated using the weights between input and hidden layer w hi (11 × 4), independent variable matrix x (4 × 1), and bias at the hidden layer b h1 (11 × 1).
The output for the first hidden layer neuron was calculated by applying log sigmoid function to the product of the first row of w hi , i.e., w 1i (1 × 4) and x (4 × 1), then added to the first row of bias at the hidden layer b h1 , i.e., b 11 (1 × 1), shown in Table 2. A similar procedure was followed for calculating the output of other neurons in the hidden layer (2 to 11) and represented in Table 2. (9)

Effect of Process Parameters on Encapsulation Efficiency
An effective encapsulation technique relies on attaining the maximum holding of the targeted compounds and the lowest amount of the core compounds on the surface of encapsulated powder molecules. The encapsulation efficiency for different combinations of process parameters varied from 56.680 to 88.935%. The high yield for Y 1 was obtained at maltodextrin of 4 g, gum arabic of 4 g, gelatin of 4 g, and ultrasound power of 150 W, whereas the lowest yield of Y 1 was found at the combination of X 1 of 3 g, X 2 of 1 g, X 3 of 3 g, and X 4 of 200 W. The plot between the experimental and predicted Y 1 of freeze-dried encapsulated product with r-value of 0.98 suggests that both data were in good agreement with each other (Fig. 3(i)). The influence of process parameters X 1 on encapsulation efficiency was studied by keeping the other three process parameters at the center level and presented in Fig. 4(i). It was observed that increasing the maltodextrin content in the wall material from 1 to 3 g resulted in a steep increase in encapsulation efficiency, and further increase to 5 g resulted in a slower rate of increase ( Fig. 4(i)). Similarly, the influence of X 2 , X 3 , and X 4 on encapsulation efficiency was studied by keeping the other three process parameters at the center level and shown in Fig. 4(ii), (iii), and (iv), respectively. The calculated and predicted relative influence of each process variable on encapsulation efficiency is shown in Table 3. The quantity of gum arabic (0.495) in the encapsulation wall material has the greatest influence on encapsulation efficiency among all four process factors, followed by the quantities of gelatin (0.332), maltodextrin (0.329), and ultrasonication power (− 0.281) ( Table 3). Comparable results were reported when maltodextrin and gum arabic were used individually as wall material for entrapping polyphenols from fermented waste water of miang processing by freezedrying encapsulation technique with 10:1 core to wall material. The encapsulated powder with gum arabic resulted in a better encapsulation efficiency (99.35%) than the encapsulated powder formed with maltodextrin (89.63%) (Muangrat et al., 2019). Similar results of a greater effect of gum arabic on encapsulation efficiency were reported during norbixin encapsulation by spray-drying using maltodextrin and gum arabic as wall materials, where the addition of gum arabic to maltodextrin increased encapsulation efficiency from 22.52 to 50.02% (Tupuna et al., 2018). Minimum encapsulation efficiency was reported while using only maltodextrin for wall material. This might be due to the typical characteristics of polymers, such as very low surface activity, the inadequacy of emulsification, and the poor ability to form films. Hence, maltodextrin is generally combined with compounds with good emulsifying capabilities (Cano-Higuita et al., 2015). Gum arabic is a highly branched heteropolymer of sugars built by a ramified carbohydrate chain and minor glycoproteins connected by covalent bonds. This favors gum arabic to combine with both hydrophilic and hydrophobic molecules in nature, and thus it is a magnificent emulsifying agent in an extensive range of pH. Hence, gum arabic is suitable with a vast range of polymers and entraps the targeted molecule because of its great ability to form films (Akhavan Mahdavi et al., 2016). The three components of blends for wall material, i.e., gum arabic, maltodextrin, and gelatin, are found to have a positive effect on Y 1 whereas ultrasound power has a negative effect. The logic behind the positive impact of wall material components can be associated with the great solubility of the component blends of targeted compounds and the low viscosity when a high composition of solids is used (Özkan & Ersus Bilek, 2014). The encapsulation efficiency Table 2 Equation for each neuron at the hidden layer of the network w hi represents the weight between input and hidden layer; b h1 bias values at hidden layer; o h is the output of the hidden layer; x 1 , x 2 , x 3 , and x 4 represent the input neuron; and suffix i, h , and o represent the neuron number for input, hidden, and output layer, respectively

Response
Response in terms of log sigmoid function logsig w hi × x + b h1 1 3 of spray-dried encapsulated product formed from the pulp of juçara (Euterpe edulis M.) using maltodextrin, gum arabic, and gelatin as wall material was found to be 83% (Bicudo et al., 2015). Gelatin mixed with other polymers for encapsulation improves the color-staining effect and thus enhances the stability of UAE extracted dragon fruit peel containing betacyanin. Similar results of improved stability of curcumin pigment were reported for spray-dried encapsulation method using gelatin and porous starch as wall material (Wang et al., 2009). The reason for the decrease in Y 1 with increase of X 4 may be associated with the thermal degradation of betacyanin with the increase of power during the mixing of the encapsulating agents and core material. This phenomenon might have resulted in a decrease in encapsulation efficiency. The heat produced in the solution during ultrasonication may be due to the collapsing bubbles, and this heat might have increased with the increase of ultrasonication power (Leong et al., 2017). A similar trend was reported during the encapsulation of coconut milk by spray drying with the application of ultrasound for homogenization, where the encapsulation efficiency was approximately 80% for ultrasonication power of 5.68 W/g, and a decrease of encapsulation efficiency of 12.5% was observed with the increase of power to 6.68 W/g but the increase of power from 2.27 to 5.56 W/g increased the encapsulation efficiency (Le & Le, 2015).

Effect of Process Parameters on Antioxidant Activity
The antioxidant activity was found to be in the range of 59.890 to 82.319% under the selected experimental conditions of four independent variables. The maximum antioxidant activity was observed when the X 1 of 4 g, X 2 of 4 g, X 3 of 4 g, and X 4 of 150 W, and the minimum antioxidant activity was found at the independent variable combination of X 1 of 3 g, X 2 of 1 g/100 mL, X 3 of 3 g, and X 4 of 200 W. The output of antioxidant activity of freeze-dried encapsulated product data obtained by the ANN model equation formed in Table 1 was plotted against the observed antioxidant data of freeze-dried dragon fruit peel encapsulated product ( r = 0.98 ) and shown in Fig. 3(ii).
From the ANN modeling, the sequence of the relative effect of process parameters was as follows: X 1 (0.695), X 2 (0.639), X 4 (− 0.458), and X 3 (0.330), shown in Table 3. Out of all the four process parameters, ultrasound power was found to have a negative effect on antioxidant activity. The three process parameters are the components of wall material found to have a positive effect, shown in Table 3. The effect of process parameters on antioxidant activity was similar to that of encapsulation efficiency of the freeze-dried encapsulated product.
The antioxidant activity of freeze-dried encapsulated powder was affected higher by X 1 compared with the other three independent variables. Similar trends of the higher effect of maltodextrin were reported during encapsulation of eggplant peel extract by spray drying method using maltodextrin and gum arabic as wall material with core to wall ratio of 1:1. The reported antioxidant activity of encapsulated egg plant peel while using maltodextrin, gum arabic, and a combination of both components as wall materials  Table 3 Relative effect of process parameters on the properties of freeze-dried encapsulated product Where in y i , "i" denotes the response neuron was 71.25%, 55.73%, and 63.52%, respectively (Sarabandi et al., 2019). Using freeze-drying method for encapsulating spent coffee grounds with maltodextrin as wall material reported antioxidant activity (FRAP assay) of 1.56 mmol Fe(II)/100 mL while the change in wall material to gum arabic, the antioxidant activity was decreased by 22.44% (Ballesteros et al., 2017). The results suggested that the use of a higher composition of X 1 in wall material was more appropriate for preserving the antioxidant activity of bioactive compounds while applying freeze-drying technique for encapsulation of bioactive compounds.
The positive effect of gum arabic on the response Y 2 of encapsulated product can be associated with the chemical structure of protein fraction in gum arabic (Tonon et al., 2010). A similar trend of increase in antioxidant activity from 84 to 89% was observed with the increase of gum arabic in core to gum arabic ratio from 1:4 to 1:6 during spray drying of propolis (Da Silva et al., 2013).
The positive effect of X 3 on the antioxidant activity may be due to the interaction with the maltodextrin at temperatures produced during ultrasonication resulting in Maillard reaction forming melanoidins. Melanoidins formed by this process tend to have high molecular weight and antioxidant activity (Wang et al., 2011). The application of ultrasound power might have affected the antioxidant property in a negative way due to the heat produced in the mixture, as the ultrasound is mainly mechanical, and the vibration of product structure caused by ultrasound application can convert this mechanical energy into heat by friction (Colucci et al., 2018). The ultrasound power has increased antioxidant activity of encapsulated product with an increase of power from 100 to 150 W, but further increase in ultrasound power demonstrated a decline in antioxidant activity, which can be visualized from Fig. 4(iv).

Effect of Process Parameters on Hygroscopicity
Hygroscopicity is chiefly influenced by inherent compositions of product, the physical structure of product like surface area and particle morphology, and also concentrations of the drying carriers and drying methods (Arshad et al., 2020;Ferrari et al., 2012). It helps in determining the shelf life and stability of the dried products where the powder is considered non hygroscopic if the hygroscopicity value is less than 10%. One approach to reduce the hygroscopicity of bioactive compound powder is by encapsulating it with high molecular weight wall materials (Bhandari et al., 2013). The high Y 3 value of 8.699 g/100 g was found for the encapsulated product that formed with wall material prepared at a power level of 250 W with the combination of X 1 , X 2 , and X 3 of 2 g, 4 g/100 mL, and 4 g, respectively.
The low Y 3 value of 6.334 g/100 g was observed for the encapsulated product formed with the wall material containing X 1 of 5 g, X 2 of 3 g, and X 3 of 3 g when homogenized at ultrasound power ( X 4 ) of 200 W. The predicted and the experimental hygroscopic data at different combinations of process parameters were found to be in good agreement with each other, suggested by a higher r-value of 0.98, and are pictorially represented in Fig. 3(iii). The influence of process parameters on the hygroscopicity of the encapsulated product is illustrated in Fig. 4. From the relative effect of process parameters, it was found that X 3 has less effect on Y 3 compared with the other three process parameters. X 1 was found to have a higher effect on the hygroscopicity; the relative effect of the four process parameters was as follows X 1 (−0.605) > X 2 (0.520) > X 4 (0.449) > X 3 (−0.434) . The X 2 and X 4 were found to have a positive effect on hygroscopicity, which means an increase of these parameters showed an increase in the Y 3 , whereas the X 1 and X 3 found to have a negative effect on the Y 3 illustrate the increase in the value of these parameters which decreased the response Y 3 . The negative effect of maltodextrin on the hygroscopicity can be attributed to its ability to enhance the stability of the dried encapsulated powder due to its less hygroscopicity nature and high molecular weight of maltodextrin (W. Wang & Zhou, 2012). The hygroscopicity of freeze-dried encapsulated product from Averrhoa carambola pomace was also reported to decrease from 6.83 to 5.55 with the increase of maltodextrin content from 10 to 20% in wall material (Saikia et al., 2015). Similar results of decrease in hygroscopicity were observed with the increase of maltodextrin from 5 to 15% while encapsulating the beetroot juice with spray drying technique (Bazaria & Kumar, 2018) and the hygroscopicity of spray-dried encapsulated product was also reported to decrease with the increase of maltodextrin (Rodríguez-Hernández et al., 2005). The increased level of X 2 , in turn, increased the hygroscopicity of encapsulated product which can be associated with easier polar interaction between its hydrogen chain and water due to different polarity and the highly branched structure of the gum arabic. A similar pattern of increase in hygroscopicity from 9.18 to 10.66% was observed with the increase of gum arabic from 0 to 2% in wall material along with maltodextrin during spray-dried encapsulation of propolis extract (Sukri et al., 2021). Results reported for microwave-assisted encapsulated anthocyanins from Ipomoea batatas showed that the product with only maltodextrin, only gum arabic, and a combination of both as wall material was found to have low hygroscopicity value of 10.028, high hygroscopicity value of 18.736, and moderate hygroscopicity value of 15.181, respectively (Mohd Nawi et al., 2015). The composition of gelatin was found to have a negative effect on the hygroscopicity of the encapsulated product, i.e., an increase in X 3 composition in the wall material decreased the hygroscopicity. Similar results of decrease in hygroscopicity from 49.05 to 45.25 g/100 g were observed with an increase of gelatin content and gum arabic content from 5 to 7 (v/v) with 50% of core material during encapsulation of anthocyanins from black raspberry by freeze-drying method (Shaddel et al., 2018).

Effect of Process Parameters on Solubility
Solubility is the most reliable parameter to assess the characteristics of dried powder in an aqueous solution, whether to form a solution or suspension in water. Higher Y 4 of 52.805% was found for the encapsulated product that formed when the X 4 was 250 W and wall material constitutes of X 1 , X 2 , and X 3 of 4 g, 4 g, and 2 g, respectively. The lowest Y 4 of 27.003% was observed for the process parameter combination of X 1 , X 2 , X 3 , and X 4 of 1 g, 3 g, 3 g, and 200 W, respectively. The effect of independent variables of the process on the dependent variable of the process is shown in Fig. 4. The observed and the ANN predicted values for Y 4 were plotted and found to have r-value of 0.99, illustrated in Fig. 3(iv).
Gum arabic composition ( X 2 ) in wall material was found to have a higher effect on solubility followed by X 1 , X 3 , and X 4 . The sequence of process variables from high to low effect is as follows: X 2 (0.773) > X 1 (0.536) > X 3 (−0.447) > X 4 (0.343) . The response Y 4 was increased with the increase of X 1 , X 2 , and X 4 which was justified by the relative effect illustrated in Table 3, while the X 3 was found to have a negative effect on the response Y 4 of freeze-dried encapsulated product. The positive effect of hydrolyzed starch such as maltodextrin on the solubility of the encapsulated product was due to lower-molecular-weight complexes resulting from the depolymerization of native starch. The increase in solubility of spray-dried encapsulated betacyanins was reported when maltodextrin was used as wall material (Cai & Corke, 2000).
The higher effect of gum arabic on solubility can be associated with high water solubility and the ability to produce low viscosity solutions at high concentrations of solids than other gums (Madene et al., 2006).
An increase of gelatin content in the wall material component influenced the solubility of encapsulated product negatively; this may be due to the formation of a coating, which delayed the leaching of low molecular weight compounds. The results were supported by the increase in solubility time of the encapsulated product of turmeric oleoresin with an increase of gelatin content in the wall material when the maltodextrin was kept constant (Malacrida & Telis, 2011). A similar trend of decrease in solubility in cold water was reported for freeze-dried microencapsulated beta-carotene with the increase of gelatin content in wall material when used along with starch (Spada et al., 2012). The low-frequency high-intensity ultrasound generates strong shear and mechanical forces (Leong et al., 2018;Yang et al., 2020). These forces might have influenced the mixing of the wall materials and core components in a positive way. In turn, proper mixing of components increased the solubility of the encapsulated product.

Optimization and Validation of Combined ANN-GA Model
The process parameters of the freeze-drying encapsulation of UAE extracted dragon fruit peel were optimized by using GA. The input or the objective function for the GA optimization was the equations formed by using the parameters of ANN model, which are weights and bias. The fitness functions of the GA are presented in Eq. (2), and this function provides a set of 18 solutions. From these solutions, the optimized condition was found to X 1 of 4.461 g, X 2 of 3.863 g, X 3 of 3.198 g, and X 4 of 123 W according to the highest fitness value using Eq. (2). The predicted values for the responses Y 1 , Y 2 , Y 3 , and Y 4 at the optimized condition were found to be 88.143%, 81.702%, 6.924 g/100 g, and 32.841%, respectively. For the validation of the proposed model, the experiment was conducted at the optimum condition X 1 of 4.5 g, X 2 of 3.9 g, X 3 of 3.2 g, and X 4 of 120 W, and the observed values for responses at this condition were found to be 89.715%, 80.039%, 7.003 g/100 g, and 32.418% for Y 1 , Y 2 , Y 3 , and Y 4 , respectively. The relative deviation between the predicted and observed responses was calculated using Eq. (8) and found to be less than 2.077%, presented in Table 4.

Thermal Stability Study of Freeze-Dried Encapsulated Dragon Fruit Peel Powder
The thermal stability of freeze-dried encapsulated dragon fruit peel powder was studied at two different pH, 7 and 3.6, at a temperature of 80 °C in terms of betacyanin content. The linear degradation of betacyanin with respect to time at two pH levels followed first order kinetic model, and the data were fitted with Eq. (6), illustrated in Fig. 5. Similarly, first order kinetic model was used to study the degradation kinetics of spray-dried encapsulated blueberry powder in terms of anthocyanins (Tao et al., 2017).
The first order kinetic model fitted the experimental data with higher R 2 (> 0.977) and lower values for errors 2 (< 9.981) and RMSE (< 2.925), presented in Table 5. The amount of betacyanin retained for the sample treated at pH 7 and 3.6 was found to be 68.485% and 60.135%, respectively. The highest retention was found for the sample treated at pH 7 with the lowest first order kinetic parameter (k) of value 0.495 ×10 −2 , min −1 . From the table, it can be observed that a decrease in pH from 7 to 3.6 increased the degradation rate constant by 35.152%. Comparable results were found where an increase in k value was observed with the decrease of pH while studying the thermal stability of microcapsule formed by gelatine and gum arabica as wall materials (Rodrigues da Cruz et al., 2019). The half-life time was calculated using Eq. (7) and presented in Table 5. The half-life time for the sample treated with pH of 3.6 was found to be 103 min, and an increase in pH to 7 increased the half-life time to 140 min. These results demonstrate that the freeze-dried encapsulated product was more stable at pH of 7 than the pH of 3.6 when treated at 80 °C temperature, as shown in Table 5. Half-life time is inversely proportional to the first order rate constant, i.e., an increase in k value decreases the t 1∕2 value.

Conclusion
In this study, the ultrasound-assisted extracted dragon fruit peel was encapsulated using the freeze-drying technique, and the process was modeled and optimized using integrated ANN-GA. The properties evaluated for the freeze-dried encapsulated powder at different combinations of process parameters were encapsulation efficiency, antioxidant activity, hygroscopicity, and solubility. At different combinations of process parameters, the encapsulation efficiency, antioxidant activity, hygroscopicity, and solubility were found to be in the range of 56.680-88.935%, 59.890-82.319%, 6.334-8.699 g/100 g, and 27.003-52.805%, respectively. The optimized concentration of maltodextrin, gum arabica, and gelatin was 6.22 g, 3.74 g, and 1.04 g, respectively. The optimum ultrasonic power was found to be 132 W. The predicted values of encapsulation efficiency (%), antioxidant activity (%), hygroscopicity (g/100 g), and solubility (%) were 88.143, 81.702, 6.924, and 32.841, respectively. The relative deviation between the predicted and experimental value at optimum conditions was found to be less than 2.077%. The data for thermal stability of encapsulated powder at 80 °C and two pH of 7 and 3.6 followed first order kinetic model, and the rate constant was 0.495 ± 0.009 and 0.669 ± 0.015 min −1 , respectively. The encapsulated powder was more thermally stable at 80 °C and pH 7 than the powder treated at pH 3.6. It was validated by half-life times of 140 and 103 min for pH 7 and 3.6, respectively. The freeze-dried encapsulated powder obtained by this approach increased heat stability and had the capacity of controlled release, and the encapsulation matrix acted as an ideal vehicle for antioxidant delivery.
Author Contribution G.V.S. Bhagya Raj and Kshirod K. Dash wrote the main manuscript text and prepared figures and tables. Kshirod K. Dash reviewed the manuscript.
Data Availability All data and materials are available upon reasonable request from the corresponding author.

Competing interests
The authors declare no competing interests.