The objective of study was extraction of pectin from orange peel using ultrasound assisted extraction and response surface method and artificial neural network technique. The accuracy of the two models was studied to compare the performances of the two models to make decision for achievement of optimum process parameters during extraction of the pectin. The following findings are absorbed from the effects of extraction parameters. The pH solution was highly significant compared to ultrasound power. As well as interaction between ultrasound and pH solution were found to be strongly influenced the extraction yield of pectin. The optimal parameters for extraction were irradiation time of 22.5 min, pH of 1.5, and ultrasound power of 155W and liquid-solid ratio 22.5:1 mL/ g. Under these conditions, yield of pectin was 26.87% experimentally, while 26.74 and 26.93% of yield were predicted by response surface methodology and artificial neural network model respectively. The extracted pectin of orange peel was categorized as high methoxyl pectin, since it has 63.13% degree of esterification, which is above 50% affirmed by Fourier transform infrared spectroscopy detection. Both response surface methodology and artificial neural network model prediction was in good agreement with experimental data; however, the prediction of artificial neural network prediction was better than artificial neural network. Therefore, artificial neural network model is much more accurate in estimating the values of pectin yield and mean square error when compared with the response surface methodology.

The orange peels were collected from juice producer located in Jimma. It was manually cleaned to remove impurity and dried for three days followed by an electric oven dry at 50°C until similar weight of the peel achieved. The dried peels were washed, mixed in a mixer grinder and sieved to avoid non-finely particles. The result powder was stored in a dark and dry out place for subsequent requirement.

_{2}O_{4}, trifluoro acetic acid, NaOH, ethanol, 2, 2-Diphenyl-1-Picrylhydrazyl] were purchased from chemical purchasers.

Experiment design was developed using Stat Ease Design Expert software version 11.1.0. Based on the design of experiment; central composite deign, the second order polynomial regression model was developed to predict the performance of variables. The extraction parameters selected according to previous literatures as shown in appendix; A

The second-order models generated by RSM are often used to determine the critical points and can be written in a general form as (Kleijnen, 2008):

Where, _{i}_{j}_{0}, β_{i}, β_{i}, β_{ii},

Using Eq.(

MATLAB® 2014a was used for the formulation of Artificial neural network model using feed forward multilayer network contains three primary layers knwon as input (used hyperbolic tangent sigmoid transfer function), hidden and output layers (used pure-linear transfer function) to anticipate the yield (Nazerian, Kamyabb, Shamsianb, Dahmardehb, & Kooshaa, 2018) as indicated in Eq.(

The coefficient of determination; R^{2}, Root Mean Squared Error; RMSE, mean average error; MAE, standard error of prediction; SEP, and absolute average deviation; AAD were determined to check the accuracy and predictive ability of ANN and RSM using Eq. (

Where; _{p}_{e}

By using previous study as reference with modification and it should also be declared that the degrees of parameters were selected depend on previous study (Hosseini

The produced pectin contents were analyzed by using prinks Elmer spectrum 65 FT-IR technique with the help of IR correlation. The wavenumber region for the analysis was 4000-400 cm^{−1} (in the mid-infrared range) and IR spectrum was reported by % transmittance.

central composite design parameters level for pectin extraction

Independent variables with experimental and response with RSM and ANN

The pectin extracted at the optimum processing conditions (the highest yield) was analyzed by determining the following properties.

_{1}). Subsequently, 15 ml of 0.4 M KOH was added, and the solution was shaken powerfully and allowed to cool down for 10 min. In addition, 15 ml of 0.6 M hydrochloric acid was added and the solution was shaken until the disappearance of pink color. The mixture was volumetrically analyzed with 0.4 M KOH for the end step and the consumed amount was collected as volume (V_{2}) and the esterification capacity was determined using Eq. (

_{1} = initial titer (ml), and _{2} = final titer (ml)

EW= Molecular mass of dried product, VA = volume of alkali (ml), and K_{en}= normality of potassium.

_{s}

A very good correspondence between the observational and anticipated values for pectin yield was obtained from the check bit plot between the anticipated and the observational values as shown in

The UAE yield of pectin was less than the result reported by, Hosseini

The regression coefficient (interms of coded), variance analysis of the expected model and all colleration coefficients( R^{2}, R^{2} adj and R^{2} pre) have been used to test the goodness of the model (Zhang, Chen, Mao, Guo, & Dai, 2014) “ISSN”: “18790003”, “abstract”: The central composite design (CCD listed in table 3. The analysis results proved that the relationship between factors and the response of the expected model was well-correlated. Residual is the difference between the actual and the fitted value of a model and it is used to predict the adequacy of the model. Small residual mean’s; model anticipated is accurate (Samavati, 2013) based on a five level, four variable central composite rotatable design (CCRD. The statistical effluences of all the terms of the model were tested by the F and P-value. The corresponding variables would be more significant if the F-value became higher and the P-value became smaller (Zhang

ANOVA for Quadratic model

Artificial neural network for input, hidden and output layers (4–_{n}

Experimental value versus RSM expected of pectin yield (%)

The Pred.R² of 0.9575 is in reasonable estimate with the Adj. R² of 0.9475; i.e. the distinction is much less than 0.2. The adequate precision ratio of 58.439 shows an adequate signal and this model can be used to navigate the design space since, adequate precision is used to find out the signal to noise ratio and a ratio greater than 4 is desirable. Meanwhile, the very small value of coefficient of variation (CV:7.23%) clearly shown a very strong degree of precision and a great deal of reliability of the experimental values (Prakash

The importance of the independent variables and their effects could be indicated by the magnitude and sign of the coefficients (Zhao-Hui

The result in

Irradiation time is one of the important process parameters that affect the yield of pectin significantly compared to L-S-R in this study. Due to creation of the cavitation bubble by ultrasound waves that supports the disruption of plant cell wall to improve the extraction efficiency of pectin and produce swelling and hydrate the plant material, in the initial stage, the extraction efficiency was increased up to 22.5 min and the reduction was observed above this value (Shivamathi

Yield of pectin did not highly influenced with liquid-solid ratio compared to other parameters; an observation that was confirmed by the results of the analysis of variance (p > 0.05) and Eq. (

Ultrasound versus pH (a) and Liquid-solid ratio versus pH on the yield of pectin (b)

The ANN anticipation has been carried out successfully using information shown in indicated in _{n}

Relative statistical indicators values of RSM and ANN model

The degree of association or relationship among the parameters in the question was specified by correlation coefficient (R). A unit (

To identify the best model that accurately predict the effect of extraction parameters on the yield, the prediction capabilities of the RSM and ANN models the computed values of all statistical indicators were compared. The result showed that both models indicates high of accurate of the result, since the higher R^{2} values were predicted in both cases. However, ANN gave a lower RMSE value when compared to the RSM model. Therefore, ANN was a better modeling tool due to its low RMSE and high extract yield. Owing to higher value of R^{2} and lower values of other statistical values of ANN compared to RSM, the accuracy of ANN model is better than that of the RSM model.

The analyzed characteristics of the extracted pectins were indicated with their values and summarized in

Properties of extracted pectin with their values.

To represent the extent to which carboxyl groups in pectin molecules exist as the methyl ester, esterification capacity (EC) is an essential molecular index for pectin classification. Since result indicates, higher degree of esterification above50 percentage, it can be considered as high methoxyl (HM) pectin. Pectin’s has high tendency to form gel rapidly at higher temperature and have a great effective action on the lipid profile, when degree of esterification is above 50 percent (Brouns

Another important characteristic in classifying the functional behavior of pectin is equivalent weight (EW) since, gelling tendency of individual pectin’s are linked very closely with EW. The higher equivalent weight, has greater gel formation, while the lower equivalent weight, indicates larger partial degradation of the pectin which is non-profitable (Hardy, 1924) strawberries, rhubarb stem and apples, Schryver and Haynies [1916] employed a hot dilute solution of ammonium oxalate to extract pectinogen from the washed and dried material. Subsequently, Farnell [1923], working in conjunction with Schryver, showed that oxalic acid could be substituted for the salt, and indeed, is preferable in so far that it remains in solution when the extracted pectinogen is precipitated by alcohol, whereas ammonium oxalate is strongly adsorbed by the alcohol-gel. Farnell also demonstrated that solutions of ammonium sulphate and of carbonic acid are capable of extracting appreciable amounts of pectinogen from dried turnip, but that water alone is without effect. He suggested that pectinogen is loosely combined with calcium in plant tissues, and is liberated therefrom by any reagent which precipitates the metal. Carre [1922] has, however, shown that dilute hydrochloric acid is an effective extractor of pectinogen, a result which is not concordant with this generalisation. In examining the factors which influence the extraction of pectinogen from the fibre of sugar cane, Farnell (private communication and thus the raise or diminish of the equivalent weight could be subjected to the content of free acid (Mohamed, 2016). The average equivalent weight of orange peel pectin in this study found to be 604.74 kDa closest to data reported by Altaf

Correlation coefficients for mean pectin yield (ANN)

The methoxyl content (MC) is an instrument tool used to control the setting time, the gel strength and to find the functional properties of pectin (Twinomuhwezi, Godswill, & Kahunde, 2020). Kanmani, (2014) established that, depending on the origin of raw material used, method of extraction, in addition to the method used for determination of methoxyl content, the MC of pectin usually varies from 0.2–12%. The result indicated that 6.23% of methoxyl content was achieved. Since the methoxyl content was below 12%, this pectin has lower ester characteristic, which implies that it is desirable in terms of quality, and in addition due to it has above 50% DE.

As indication of Food Chemical Codex (FCC), Food and Agriculture Organization (FAO), and European Union (EU), pectin must contains at least 65% of galacturonic acid (Willats, Knox, & Mikkelsen, 2006), since anhydrouronic acid amount is used to represent the gelling capabilities of given pectin. The higher value means, the produced pectin has a lower amount of protein. In present study, the result indicated that 68.25% of anhydrouronic acid was achieved. A minimum value of anhydrouronic acid (65%) for commercial pectins has been specified by FAO (Twinomuhwezi

Acetyl value (AV); the gelling capacity of pectin diminished with the raise in the degree of acetylation (Fakayode & Abobi, 2018). The inhibition of gelormation will be created when acetyl formation is found in pectin. Other researcher shown that the gelling capacity of pectin diminished with raise in the degree of acetylation and samples holding 3.5%-4.0% acetyl brings weak gels while gelling capacity restored at levels around 2.4% acetyl (Mohamed, 2016). Based on acetyl value and compared to previous study, the result indicates that orange peel pectin has good gelling capacity (0.371%) indicated in

The ultrasound-assisted extraction of pectin was optimized by the combination of both response surface methodology (RSM) and artificial neural network multi-layer back propagation (ANNMBP). The objectives of study were extraction of pectin from orange peel using ultrasound assisted extraction and response surface method and artificial neural network technique. The accuracy of the two models were studied to compare the performances of the two models to making decision for achievement of optimum process parameters during extraction of the pectin.

The extracted pectin of orange peel was categorized as high methoxyl pectin since it has 63.13% degree of esterification that is above 50% affirmed by FTIR detection.

The prediction of yield was investigated by using RSM and ANN model and found to be in good agreement with experimental data, however, the prediction of ANN model is found to be better than RSM. The results indicates that, ANN model is much more accurate in estimating the values of pectin yield and mean square error when compared with the RSM.

I like to acknowledge Jimma institute of Technology, School of Chemical Engineering, Addis Ababa University, Department of Chemistry and all lab technicians for their support in different ways.

Irradiation time

Analysis of Variance

pH: power of hydrogen

Ultrasound power

Coefficient of variation

Liquid to solid ratio

Potassium hydroxide

^{2}

Coefficient of determination

Response surface method

Standard deviation

Fourier Transform Infrared spectroscopy

Artificial neural network

Watt

_{2}: Neural network and adaptive neuro fuzzy interface system modeling and response surface optimization