Response Surface Methodology and Artificial Neural Network

Techniques for Prediction and Optimization of Biodiesel Yield

Synthesized from Cottonseed Oil
Negasa Tesfaye Tefera1*, Ramesh Babu Nallamothu1, Getachew Alemayehu Lakew1, Teshome Kumsa Kurse2
1,2
Department of Mechanical Engineering, Adama Science and Technology University, Adama, Ethiopia

*Author for correspondence: Email: [email protected]

Abstract
In this study prediction and optimization of biodiesel yield synthesized from cottonseed oil were carried out using response
surface methodology (RSM) and artificial neural network (ANN) techniques. Box-Behnken Design (BBD) of RSM was used to
create the experimental design matrix. The ANN model was established with three input layers, ten hidden layers, and one output
neuron layer (3-10-1). Three process variables were reaction time (40, 60, and 80 min.), catalyst concentration (0.5, 1, and 1.5wt.
%), and methanol-to-oil molar ratio (1:4, 1:6, and 1:8). The findings of the experiment demonstrated that biodiesel produced
with a catalyst of 1 wt. %, a reaction duration of 60 min., and a molar ratio 1:6 have a maximum yield of 94.66%. The findings
of the experiments were contrasted with the predictions made by the RSM and ANN models. The biodiesel yield was represented
as a quadratic model from the RSM analysis, with a determination coefficient (R2) and adjusted R2 of 0.953 and 0.95, respectively.
The Levenberg-Marquardt approach was used in ANN modeling to train model networks for biodiesel yield prediction. At epoch
3, the observed mean squared error (MSE) was 4.963e-18, and the observed R-value was 0.9957. Gas chromatography–mass
spectroscopy (GC-MS) analysis was used to determine the fatty acid methyl ester (FAME) concentration in the biodiesel.
Furthermore, the important physicochemical characteristics of the biodiesel comply with the specifications set forth by the Amer-
ican Society of Testing and Material (ASTM) D6751 and European norms (EN) 14214 standards.
Keywords: Optimization, Response surface methodology, Artificial neural network, Biodiesel, Gas chromatography–mass
spectroscopy

1. Introduction
Global energy demands are growing daily because of population expansion, industrialization, and technological advancements.
Approximately 80% of the global energy requirements might be satisfied by petroleum-based fuels and 20% by renewable energy
[1]. Climate change, global warming, and environmental harm are mostly caused by a continuing rise in emissions from the
combustion of fossil fuels [2]. However, the necessity for energy security and environmental preservation becomes imperative
due to limited fossil fuel supplies, environmental degradation, and fluctuating fuel prices. In other words, it is necessary to search
for alternate sources, particularly oil seeds [3]. Oil from raw vegetable sources is the subject of recent technological advance-
ments in both domestic and commercial applications. This is explained by their suitability as bio-oil for the industrial production
of products such as biodiesel, bio-lubricants, bio-transformers fluids, as biodiesel can serve as an alternative to petroleum-based
fuels [4]. Biodiesel (methyl ester) is, therefore, considered a viable alternative renewable fuel to reduce the problems associated
with energy and air pollution [5,6]. Biodiesel is the most promising option for widespread implementation with indigenous raw
materials and a degree of technological maturity and production readiness that satisfies essential biofuel requirements [7]. Owing
due to its enhanced fuel qualities such as its lubricity, high cetane index, inherent molecular, and minimal sulfur content, biodiesel
is a viable substitute for conventional petro-diesel fuel [8–10]. Owing to its poor atomization and high viscosity (> 40 cSt),
biodiesel is not recommended for use in diesel engines [11,12]. The transesterification method was used to lower the viscosity,
in addition to thermal cracking, micro-emulsification, and blending techniques [13,14]. The transesterification method has at-
tracted considerable interest because of its high biodiesel conversion efficiency. The triglycerides in raw oils react with alcohol
and catalysts during the process of transesterification to produce methyl or ethyl esters. Methanol produced reduced viscosity
compared to ethanol [7]. A catalyst is used to accelerate the reaction rate and improve the quality of the methyl ester [15]. Base
catalysts are preferred over acid catalysts because they can complete the reaction more quickly, have a greater conversion effi-
ciency, and require a lower reaction temperature [16]. As a base catalyst, KOH has the following characteristics: it reacts at low
temperature, takes little time to react, works at atmospheric pressure, generates no intermediate products, and converts raw
materials at a high rate [7]. Numerous investigations have been carried out to optimize biodiesel production from different types
of feedstocks. Among those studies, RSM and Taguchi were the most frequently used methods for methyl ester optimization.
Despite the differences in methodologies and challenges encountered, all approaches aim to optimize the requirements for pro-
ducing biodiesel [1]. Özgür C. [17] utilized a sodium hydroxide catalyst to optimize the synthesis of biodiesel from pomegranate
seed oil. The individual and combined effects of the process factors were investigated using a central composite design (CCD).
The investigation revealed that a reaction temperature of 70 °C, a methanol-to-oil ratio of 1:8.01, catalyst concentration of 1.08
wt.%, and a duration of 45 minutes produced a maximum yield of 95.68%. The most significant effect on the yield of methyl
esters was the methanol to oil ratio. In addition, with an R2 value of 0.9887, the RSM model provided an excellent prediction of
the experimental results. Yussuf et al. [18] developed a barium-modified zeolite catalyst for the production of methyl esters from
waste cooking using a co-precipitation method and thermal treatment. CCD was used to optimize the biodiesel process factors.
According to the study, a reaction temperature of 65° C, a 1:12 molar ratio, and 3 wt.% catalyst concentration were all necessary
to obtain a 92% maximum yield. Jimoh et al. [19] optimized biodiesel produced from used cottonseed oil (CSO) using RSM.
The studies found the optimum yield of biodiesel (93.60%) at a 1:10 methanol to oil ratio, 80 min of reaction duration, a reaction
temperature of 60oC, and 2.5 wt.% catalyst loading. Abdullah et al. [8] produced biodiesel from Allamanda seed oil (ASO). The
methyl ester process variables were optimized using the Box-Behnken Design (BBD). The investigation showed maximum bio-
diesel yield of 90.67% was obtained at 52.5 oC, 180 min reaction time, 0.5wt.% concentration of catalyst and fixed methanol to
ASO ratio of 5:1. Razzaq et al. [20] Produced the biodiesel from a palm oil and CSO mixture via an ultrasound-assisted trans-
esterification process. The maximum biodiesel (P60C40) yield predicted using response surface methodology was 96.41%. The
optimum operating variables were methanol to oil ratio (47.5 %v/v), time (35 min), and catalyst (1wt.%). They revealed a strong
correlation between the predicted and experimental findings.
Among various optimization techniques, artificial neural network (ANN) and RSM are reliable due to their accuracy and minimal
error [1]. Vinoth et al. [21] The process factors were optimized using the ANN and RSM approaches. Levenberg-Marquardt
analysis was performed on the backpropagation network. According to the study, the R 2 values for ANN and RSM were deter-
mined to be 0.97 and 0.87, respectively. Garg et al. [22] utilized ANN and RSM techniques to optimize biodiesel synthesis from
algal oils. According to the study's findings, ANN is more predictable than RSM. For both ANN and RSM, a significant quadratic
regression model was found, with R2 values of 0.99 and 0.96, respectively. Ayoola et al. [23] compared the relationship between
the process variables and the yield of methyl esters from waste soybean oil by applying ANN and RSM optimization methods.
They discovered that the ANN and RSM have determination coefficients (R2) of 0.98 and 0.93, respectively. Soj-Adekunle [24]
Investigated RSM and ANN modeling techniques to forecast the biodiesel yield from waste cooking. They computed the average
absolute deviation (AAD) and coefficient of determination values (R 2). They discovered that for ANN the AAD was 0.4930
while R2 was 0.9950 and for RSM the AAD and R2 were 0.9376 and 0.9843, respectively.

This study aims to apply RSM based on a Box-Behnken Design and ANN to predict and optimize the factors that affect cotton-
seed oil biodiesel production and explore the effects of production factors on biodiesel yield. The factors investigated were the
reaction duration, catalyst concentration, and methanol-to-oil ratio. Additionally, the accuracy and generalization capacity of the
RSM and ANN predictions were evaluated by comparing them with experimental values.

2. Materials and Methods

2.1 Materials Used
Cottonseed oil was obtained from Adama, Ethiopia. Potassium hydroxide catalyst (KOH), methanol (CH 3OH, 99.5% purity),
ethanol (C2H5OH), sodium hydroxide (NaOH), hydrochloric acid (HCl), chloroform, phenolphthalein, sodium thiosulphate, and
starch soluble were used. During the biodiesel production and characterization, all chemicals used were analytically pure.
2.2 Biodiesel Synthesis Process
Biodiesel was synthesized using the transesterification method. In this method, an accurate quantity of the cottonseed oil is
measured and added to a beaker. The oil was then preheated to 60 °C. Several alcohol molar ratios and catalyst quantities were
used to prepare the potassium methoxide solution. The cottonseed oil to 60 oc was mixed with this solution. The transesterification
reaction was conducted using various catalyst concentrations, molar ratios, and reaction durations. After the transesterification
reaction was completed, the products were put into a separating funnel and left to be suspended for 24 h. After extracting the
crude glycerol from the methyl ester, the product was repeatedly washed and heated to 130 °C for an hour to eliminate water and
the leftover residue. Finally, Equation (1) was used to determine the biodiesel yield [8,20].
volume of biodiesel obtained
% Biodiesel yield = × 100 (1)
Volume of oil comsumed

2.3 Biodiesel Characterization

Biodiesel was synthesized from cottonseed oil using KOH as a catalyst. The European and ASTM standards were applied to the
characterization of biodiesel. Biodiesel properties such as density, kinematic viscosity, iodine value, saponification value, acid
value, flash point, cloud point, and moisture content, were experimentally determined. The type and percentage of fatty acid
methyl esters (FAMEs) present in the biodiesel produced from CSO were determined using a GC-MS instrument (Agilent,
5977C, GC/MSD).
2.4 Design of Experiment Using Response Surface Methodology
The response matrix was examined, and the best parameter combination was determined, using a Box-Behnken Design with
three variables. The cottonseed oil biodiesel yield percentage was mostly influenced by process factors which are independent
variables. There was an enormous effect on the yield percentage of methyl ester when all three factors were changed. Optimizing
the parameters can help to save time and energy utilization, resulting in the highest rate of biodiesel yield with the least amount
of time and energy wasted. Consequently, RSM method has been used to increase the biodiesel yield from cottonseed oil. For
yield optimization, a total of 15 runs were observed. Table 1 lists the operational parameter ranges for the transesterification
process. Equation (2) was applied to determine the biodiesel yield by adjusting the operating conditions [20].

𝑌 = 𝑋0 + ∑𝑘𝑖=1 𝑋𝑖 𝐴𝑖 + ∑𝑘𝑖=1 𝑋𝑖𝑖 𝐵2 𝑖 + ∑𝑘𝑗=𝑖+1. ∑𝑘𝑖=1 𝑋𝑖𝑗 𝐶𝑖𝑗 (2)

Where Y is biodiesel yield; 𝐴𝑖 , 𝐵𝑖 and 𝐶𝑖 are the input factors (independent variables); 𝑋0 and 𝑋𝑖 represent intercept and first-
order regression coefficient, respectively; k is total number process factors and 𝑋𝑖𝑖 is quadratic regression coefficient.
TABLE 1: Range of input process factors for biodiesel yield optimization

Input factors Unit Coded Low Average High

Reaction time minute A 40 60 80
Catalyst concentration wt.% B 0.5 1 1.5
Methanol-to-oil ratio mole/mole C 1:4 1:6 1:8

2.5 Artificial Neural Network (ANN) Modelling

Uncontrollable variables cannot be included in the RSM, which is a limitation. Artificial neural network (ANN) an essential
component of machine learning tools that represent intricate correlations in data and generate predictions and decisions. Synaptic
weight connections are present in ANN neurons, and these connections allow neurons to store information [25]. The reaction
time, catalyst concentration, and methanol-to-oil ratio were employed as independent input variables in a combined data set that
included 15 samples. Using TRAINLM as a training function, LEARNGDM as an adaptation learning function, and MSE as a
performance function, a feed-forward backprop network was chosen for this investigation. Using the Levenberg-Marquardt for
training, an ANN model consisting of three input layers, ten hidden layers, and one output layer was examined (see Figure 1).
The TANSIG transfer function was included in these layers. The biodiesel yield (output response) in the designed model was the
outcome of 15 experiments of input factors. The ANN was executed on the R2020a version of MATLAB software. Performance
indicators such as mean square error (MSE), coefficient of correlation (R2), and root mean square error (RMSE) were utilized to
assess the applicability and accuracy of the RSM and ANN models. The following three equations define these evaluation met-
rics, which offer numerical assessments of the accuracy and predictability of the RSM and ANN models [26].
∑n
i=1(𝑋𝑒 −𝑋𝑝 )
2
MSE = (3)
n
1
1
RMSE = ( ∑ni=1(𝑋𝑒 − 𝑋𝑃𝑖 )2 )2 (4)
n

∑n (𝑋𝑒 −𝑋𝑝𝑖 )2
R2 = 1 − ∑ni=1 2 (5)
i=1(𝑋𝑒𝑖 −𝑋𝑎𝑣 )

Where n is the total number of experiments; 𝑋𝑒 and 𝑋𝑝 are the observed value and predicted value, respectively and 𝑋𝑎𝑣 is an
average value.

FIGURE 1: ANN architecture for biodiesel yield

3. Results and Discussion
3.1 Physicochemical Characterization of Biodiesel
Table 2 displays the results of the physicochemical properties of biodiesel with the EN 14214 and ASTM D6751 biodiesel stand-
ards. From Table 2, it was observed the density and viscosity of biodiesel were 0.88 g/ml at 20 oC and 4.4 mm2/s at 40oC, respec-
tively. This means that both properties satisfy the EN 14214 and ASTM D6751 biodiesel standards. The flash and cloud points
of the synthesized biodiesel were 178 oC and 5 oC, respectively. Both were within the limits of ASTM D6751. The maximum
acid value and iodine value allowed by the EN 14214 standard for biodiesel are 0.5 mg KOH/g and 120gI/100g respectively. The
acid value and iodine value produced by biodiesel in this study were 0.22 mg KOH/g and 82.2 gI/100g, respectively. Neither the
EN 14214 nor ASTM D6751 specifications specify the saponification value of biodiesel. The saponification value was 137.45
mg KOH/g. The maximum moisture percentage of biodiesel according to ASTM D6751 standard is 0.05 %. The moisture content
of biodiesel was 0.04%, which was within the limit of ASTM D6751's limitation.
TABLE 2: Physicochemical properties of biodiesel in comparison with diesel fuel
Properties Unit Diesel Biodiesel Standards
ASTM D 6751 EN 14214
Viscosity at 40 oC mm2/s 3.15 4.4 1.9-6 3.5-5
o
Density at 20 C g/ml 0.83-0.86 0.88 - 0.86-0.9
Acid value mgKOH/g - 0.22 0.5 max 0.5 max
Saponification value mg KOH/g - 137.45 - -
Iodine value g I/100 g - 82.2 - 120
o
Flash point C 69 178 130 min >101
o
Cloud point C Max +5 +5 -3 to 12 -
Moisture content % - 0.04 0.05 max -

3.2 GS-MS Analysis of Biodiesel

The fatty acid methyl esters (FAMEs) profile of cottonseed oil biodiesel determined using GC-MS equipment is shown in Table
3. The main types of fatty acids found in biodiesel are linoleic, palmitic, myristic, and oleic acids. CSO biodiesel contained more
unsaturated fatty acids than saturated fatty acids overall. GC-MS analysis revealed that the biodiesel produced contained different
compounds, including methyl esters, sterol, tocopherol, terpene, and aldehydes. The spectral distributions of palmitic acid and
linoleic acid, which have higher percentages of synthesized biodiesel, are depicted in Figure 2 and Figure 3, respectively.
TABLE 3: Weight percentage of fatty acid found in synthesized biodiesel
Chemical formula Compound Designation Weight (%)
C19H34O2 Linoleic acid C18:2 51.15%
C17H34O2 Palmitic acid C16:0 24.77%
C19H36O2 Oleic acid C18:1 14.4%
C15H30O2 Myristic acid C14:0 2.6%
C23H46O2 Behenic acid C22:0 0.81%
C31H64 Hentriacontane acid C31:0 0.98%
C25H50O2 Tetracosanoic acid C24:0 0.56
C17H32O2 Palmitoleic acid C16:0 1.8%
C9H18O2 Octanoic acid C8:0 0.1%
Others - - 2.83
 FIGURE 2: Mass spectrum distribution of Linoleic acid

FIGURE 3: Mass spectrum distribution of Palmitic acid

3.3 Biodiesel optimization using RSM
Because biodiesel production is mostly dependent on process parameters, RSM creates an interaction between them during the
transesterification process. At the optimal operating parameters, the biodiesel yield was the highest [20]. The Box-Behnken
Design of RSM was used to develop a total of 15 experiments. The three variables under investigation were the reaction duration
(A), concentration of the catalyst (B), and ratio of methanol to oil (C). RSM was used to analyze the impact of these factors on
the biodiesel yield. Table 4 shows the experimental matrix of the biodiesel yield values, along with values of the factors. Equation
(6) shows the association between the factors under investigation and the biodiesel yield.
% Biodiesel Yield = 94.66 − 1.703A − 0.217B − 4.4C − 1.71A × A − 9.14B × B − 5.02C × C − 0.14A × B − 0.08A × C +
1.62B × C (6)
In the second-order polynomial equation, the positive and negative signs represents antagonistic and synergistic effects, respec-
tively [20]. The P-value and F-value were used to describe the regression coefficients as well as the significance of the response
quadratic model for the production of biodiesel. The analysis of variance (ANOVA) results are shown in Table 5. ANOVA values
were considered significant when the F-value was larger and the p-value was smaller (P-value<0.05) [27]. The P-value obtained
via ANOVA was 0.008, indicating that the model was significant. Among model terms, C, B 2, C2, and AB were significant while
A, B, AB, BC, and AC were not significant or had less influence on the biodiesel yield compared to other model terms. The
response efficiency is determined by the F-value from analysis of variance, with values of 11.36 for a model, 4.17 for reaction
time (A), 27.85 for the molar ratio (C), 55.43 for B2, and 16.75 C2. The high F-value for A and C suggests that these experimental
factors have a greater impact on the conversion of cottonseed oil to biodiesel. The R 2 and adjusted R2 were analyzed to assess
the competency of the model. If the Adj–R2 value is more corresponding to the relevant R2 value and the R2 value is closer to
unity, indicates an adequate model fit [28]. In this study, high R2 (0.953) and adj R2 (0.95) results indicated that the model was
suitable for predicting an accurate relationship between the response and significant factors. The ANOVA results for the output
variable show that the actual and predicted values have a strong correlation, as shown in the Figure 4 with, an extremely high R2
value of 0.953.

TABLE 4: Experimental matrix with the yield

Run Factors Results
A B C Yield (%)
Coded Actual Coded Actual Coded Actual Experimental RSM ANN
1 -1 40 0 1 1 1:8 86.00 85.7087 85.6195
2 0 60 0 1 0 1:6 94.66 94.66 94.65
3 -1 40 0 1 -1 1:4 93.62 93.8612 93.6168
4 1 80 0 1 1 1:8 82.50 82.2587 83.2133
5 -1 40 -1 0.5 0 1:6 87.33 85.9812 87.259
6 0 60 0 1 0 1:6 94.66 94.66 94.65
7 0 60 1 1.5 -1 1:4 85.50 84.86 85.4403
8 0 60 -1 0.5 -1 1:4 85.75 86.8575 85.7602
9 -1 40 1 1.5 0 1:6 83.75 85.148 84.7751
10 0 60 0 1 0 1:6 94.66 94.66 94.65
11 0 60 1 1.5 1 1:8 78.75 77.6425 78.75
12 1 80 1 1.5 0 1:6 80.00 81.3487 80.0443
13 0 60 -1 0.5 1 1:8 75.25 76.89 76.8679
14 1 80 -1 0.5 0 1:6 84.16 83.7612 83.8651
15 1 80 0 1 -1 1:4 90.00 90.2912 90.622
TABLE 5: ANOVA results of the model

Source DF Adj SS Adj MS F-Value P-Value

Model 9 568.541 63.171 11.36 0.008
Linear 3 178.447 59.482 10.69 0.013
A-Reaction time 1 23.188 23.188 4.17 0.097
B -Catalyst concentration 1 0.378 0.378 0.07 0.805
C-Molar ratio 1 154.880 154.880 27.85 0.003
A2 1 10.828 10.828 1.95 0.222
B2 1 308.285 308.285 55.43 0.001
C2 1 93.140 93.140 16.75 0.009
AB 1 0.084 0.084 0.02 0.907
AC 1 0.022 0.022 0.00 0.952
BC 1 10.563 10.563 1.90 0.227
Error 5 27.810 5.562
Lack-of-Fit 3 27.810 9.270
Pure Error 2 0.000 0.000
Total 14 596.350
 FIGURE 4: Predicted vs actual value of biodiesel yield
The interactive effects depicted in Figure 5 and Figure 6 examine the interaction effect of a significant combination of process
variables on biodiesel production, providing ta response surface plot. The relationship between reaction time and catalyst con-
centration at a constant methanol-to-oil molar ratio (1:6) is illustrated in Figure 5. The results showed that more cottonseed oil
was converted to biodiesel until the optimal yield was reached as the catalyst concentration and reaction duration increased.
However, when the reaction time and catalyst concentrations were increased, the biodiesel yield decreased. The reason for this
observation is that beyond the optimal catalyst concentration and reaction time, biodiesel yield is decreased due to prolonged
reaction time can lead to reversible or side reactions and higher catalyst concentration causes the formation of soap during the
transesterification reaction [19]. When the catalyst concentration (1wt.%) was held constant, the overall effects of the reaction
duration and molar ratio on biodiesel yield are shown in Figure 6 as a surface plot. The results indicated that the yield of methyl
ester increased with reaction duration and molar ratio, but the yield of biodiesel decreased with further increases in both param-
eters. Therefore, the highest yield of biodiesel (94.66%) was achieved at a molar ratio of 1:6, with a duration of 60 min, and with
catalyst concentration of 1 wt.%. The reason for this observation is that beyond the optimal time and molar ratio, the biodiesel
yield decreases because longer reaction times might result in reversible side reactions, and inadequate methanol can cause in-
complete transesterification [19].

FIGURE 5: Surface plot of yield vs reaction time and catalyst

 FIGURE 6: Surface plot of yield vs reaction time and molar ratio
3.4 Biodiesel yield optimization using ANN
In ANN modeling, choosing the correct training function, number of neurons, and network type is crucial for forecasting the
experimental results. Further applications of ANN include 2nd-order polynomial regression derived from the response surface
methodology model[20]. In this investigation, the ANN model was established with a structure of three input layers, ten hidden
layers, and one output neuron layer (3-10-1). The reaction time, catalyst concentration, and molar ratio are represented by the
first three neurons, while the biodiesel yield is represented by the output layer (see Figure 7). In ANN modeling, the accuracy of
the model was determined by R values [17]. The training, validation, and test R values were 0.9934, 1, and 1, respectively. In
previous study Razzaq et al., [20] predicted the yield of biodiesel derived from a mixture of palm and cottonseed oil. They
observed R values of training, validation, and test were 0.967, 1, and 0.996, respectively. The validation outputs vs target plots
produced a linear regression Y = 1∗T + 5.5e − 09 to forecast the values of the ANN model. In this case, Y and T denote the ANN
model's values and the experimental values utilized to generate the related ANN values, respectively. Overall, there was a good
agreement between the estimated and experimental values, as evidenced by the R-values for these experiments. The mean squared
error (MSE) was observed at the best validation performance of 4.963e-18 at epoch 3. Figure 8 and Figure 9 show the artificial
neural network model performance and regression analysis, respectively.

FIGURE 7: Input layer, hidden layer, and output layer of the artificial neural network model
 FIGURE 8: Artificial neural network model performance

FIGURE 9: Artificial Neural Network model regression analysis

3.5 Comparison analysis of RSM and ANN
The regression factor and error values were used to calculate the model of response surface methodology and its prediction
capabilities [29]. The coefficient of determination (R2), mean square error (MSE), and root mean square error (RMSE) values
for the RSM and ANN models are displayed in Table 6. The R2, MSE, and RSME were determined using equations (3), (4) and
(5), respectively. The values of R2, MSE, and RMSE using the ANN were 0.9957, 0.384, and 0.590 respectively. In contrast, the
R2, MSE, and RMSE values obtained using RSM were 0.953, 0.769, and 0.877, respectively. By comparing the ANN and exper-
imental results, with those of the RSM results and experimental results, it shown that the biodiesel yield was predicted more
accurately by ANN than by RSM., because its MSE and RMSE values, were lower and R 2 is higher. In a previous study Ahmad
et al., [30] compared RSM and ANN modeling for biodiesel production from waste cooking oil. Three statistical parameters (R 2,
MSE, and RMSE) were used to determine the most suitable model. The ANN showed a higher R 2 (0.993), lower MSE (0.072),
and RMSE (0.268) compared to RSM (0.986), (0.081), and (0.285), respectively. In another investigation by Muhammad et al.,
[25] produced and optimized biodiesel from wet microalgae, a higher R 2 (0.94) and lower RMSE (0.38) were obtained when
ANN and RSM were used for optimization. This investigation also reported that ANN had better predictive capability than RSM.
Figure 10 compares the biodiesel yield obtained experimentally with that obtained using the RSM and ANN models.
TABLE 6: RSM and ANN model comparison statistically
Parameter RSM ANN
R2 0.953 0.9957
MSE 0.769 0.348
RMSE 0.877 0.590

Exprimental Value RSM Predicted ANN Predicted

100

90

80

70
Yield (%)

60

50

40

30

20

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Experiment Run

FIGURE 10: Comparison of experimentally yield with RSM and ANN model

Conclusions
This study describes a transesterification reaction used to synthesize and optimize biodiesel derived from cottonseed oil (CSO)
using a KOH catalyst. Three factors (reaction duration, catalyst concentration, and methanol-to-oil ratio) were used to determine
the optimum conditions. Minitab 20.3 and MATLAB R2020a software were used to modeling and predict the biodiesel yield.
The highest yield of biodiesel 94.66% was predicted using RSM and 94.65% using ANN under operating conditions of reaction
duration (60 min.), catalyst concentration (1wt.%), and molar ratio (1:6) while the biodiesel yield experimentally obtained was
94.66%. The values of R2, MSE, and RMSE using the ANN were 0.9957, 0.384, and 0.590 respectively. In contrast, the R2, MSE,
and RMSE values obtained using the RSM were 0.953, 0.769, and 0.877, respectively. Compared to RSM, ANN demonstrated
high precision and effectiveness. Owing to their high correlation concentration, both the models showed good agreement with
the experimental findings. It was discovered that ANN was more efficient than RSM in terms of R 2, MSE, and RMSE. These
findings imply that the artificial neural network has a superior predictive capacity for to maximizing the production of biodiesel
from cottonseed oil. To determine the FAME content of CSO biodiesel, GC-MS analysis was performed. GC-MS analysis was
also used to assess the saturated and unsaturated fatty acid contents. The main types of fatty acids found in biodiesel are linoleic,
palmitic, myristic, and oleic acids. Analyzing the most important fuel properties of the synthesized methyl ester, it was discovered
that its viscosity, density, cloud point, acid value, flash point, moisture content, iodine value, and saponification value all matched
ASTM D6751 and EN 14214 standards.

Data Availability
The data used to support the findings of this study are included within the article.
Conflicts of interest
The authors declare that there is no conflict of interest.

Author Contributions
Negasa Tesfaye: Conceptualization, Data curation, Formal analysis, and Writing-original draft.
Ramesh Babu Nallamothu: Data curation, Formal analysis, Investigation, and Editing.
Getachew Alemayehu: Conceptualization, Methodology, Supervision and Visualization
Teshome Kumsa: Resources, Software and Validation
Funding Statement
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Acknowledgments
The authors would like to thank the department of Chemistry of Adama Science md Technology University for providing the
material and equipments used in this study.

References
[1] Dharmalingam B, Balamurugan S, Wetwatana U, Tongnan V, Sekhar C, Paramasivam B, et al. Comparison of neural
network and response surface methodology techniques on optimization of biodiesel production from mixed waste cook-
ing oil using heterogeneous biocatalyst. Fuel 2023; 340:127503. https://doi.org/10.1016/j.fuel.2023.127503.

[2] Agu CM, Menkiti MC, Nwabanne JT, Onukwuli OD. Comparative assessment of chemically modified Terminalia
catappa L. kernel oil samples–a promising ecofriendly transformer fluid. Ind Crops Prod 2019; 140:111727.
https://doi.org/10.1016/j.indcrop.2019.111727.

[3] Agu CM, Orakwue CC, Menkiti MC, Agulanna AC, Akaeme FC. RSM/ANN based modeling of methyl esters yield
from Anacardium occidentale kernel oil by transesterification, for possible application as transformer fluid. Current Re-
search in Green and Sustainable Chemistry 2022; 5:100255. https://doi.org/10.1016/j.crgsc.2021.100255.

[4] Agu CM, Kadurumba CH, Agulanna AC, Aneke OO, Agu IE, Eneh JN. Nonlinear Kinetics, Thermodynamics, and par-
ametric studies of Colocynthis vulgaris Shrad seeds oil extraction. Ind Crops Prod 2018; 123:386–400.
https://doi.org/10.1016/j.indcrop.2018.06.074.

[5] Dehghani Madvar M, Aslani A, Ahmadi MH, Karbalaie Ghomi NS. Current status and future forecasting of biofuels
technology development. Int J Energy Res 2019; 43:1142–60. https://doi.org/10.1002/er.4344.

[6] Mahlia TMI, Syazmi Z, Mofijur M, Abas AEP, Bilad MR, Ong HC, et al. Patent landscape review on biodiesel produc-
tion: Technology updates. Renewable and Sustainable Energy Reviews 2020; 118:109526.
https://doi.org/10.1016/j.rser.2019.109526.

[7] Kolakoti A, Setiyo M, Rochman ML. A Green Heterogeneous Catalyst Production and Characterization for Biodiesel
Production using RSM and ANN Approach. International Journal of Renewable Energy Development 2022; 11:703–12.
https://doi.org/10.14710/ijred.2022.43627.

[8] Abdullahi K, Ojonugwa SS, Yusuff AS, Umaru M, Mohammed IA, Olutoye MA, et al. Optimization of biodiesel pro-
duction from Allamanda Seed Oil using design of experiment. Fuel Communications 2023; 14:100081.
https://doi.org/10.1016/j.jfueco.2022.100081.
[9] Azizzadeh Hajlari S, Najafi B, Faizollahzadeh Ardabili S. Castor oil, a source for biodiesel production and its impact on
the diesel engine performance. Renewable Energy Focus 2019; 28:1–10. https://doi.org/10.1016/j.ref.2018.09.006.

[10] Malla FA, Bandh SA, Wani SA, Hoang AT, Sofi NA. Biofuels: Potential Alternatives to Fossil Fuels. Biofuels in Circular
Economy, Springer; 2023, p. 1–15.

[11] KOLAKOTİ A. Optimization of biodiesel production from waste cooking sunflower oil by taguchi and ann techniques.
Journal of Thermal Engineering 2020;6:712–23. https://doi.org/10.18186/thermal.796761.

[12] Supriyadi S, Purwanto P, Anggoro DD, Hermawan H. The effects of sodium hydroxide (naoh) concentration and reaction
temperature on the properties of biodiesel from Philippine Tung (Reutealis Trisperma) Seeds. Automotive Experiences
2022; 5:57–67. https://doi.org/10.31603/ae.5986.

[13] Hariyanto RAB, Firmansyah RA, Burhan RYP, Zetra Y. Synthesis of bio-additive for low sulphur diesel: Transesterifi-
cation of soybean oil and ethylene glycol using K2CO3 catalyst. Automotive Experiences 2021; 4:44–50.
https://doi.org/10.31603/ae.4694.

[14] Zetra Y, Sholihah SMW, Burhan RYP, Firmansyah RA. Synthesis and characterization of diesel lubricity enhancer
through transesterification reaction of palm oil with 1, 2-ethanediol. Automotive Experiences 2021; 4:104–11.
https://doi.org/10.31603/ae.4664.

[15] Maheshwari P, Haider MB, Yusuf M, Klemeš JJ, Bokhari A, Beg M, et al. A review on latest trends in cleaner biodiesel
production: Role of feedstock, production methods, and catalysts. J Clean Prod 2022;355. https://doi.org/10.1016/j.jcle-
pro.2022.131588.

[16] Changmai B, Vanlalveni C, Ingle AP, Bhagat R, Rokhum L. Widely used catalysts in biodiesel production: A review.
RSC Adv 2020; 10:41625–79. https://doi.org/10.1039/d0ra07931f.

[17] ÖZGÜR C. Prediction and optimization of biodiesel production by using ANN and RSM. International Journal of Auto-
motive Engineering and Technologies 2022; 11:53–62. https://doi.org/10.18245/ijaet.1057170.

[18] Yusuff AS, Bhonsle AK, Bangwal DP, Atray N. Development of a barium-modified zeolite catalyst for biodiesel produc-
tion from waste frying oil: Process optimization by design of experiment. Renew Energy 2021; 177:1253–64.
https://doi.org/10.1016/j.renene.2021.06.039.

[19] Jimoh A, Agbaji EB, Ajibola VO, Uba S. Production of Methyl Ester from Used Cottonseed Oil Optimized Using Box
Behnken Design Approach. Advanced Journal of Chemistry, Section A 2023;6:1–16.
https://doi.org/10.22034/ajca.2023.359977.1327.

[20] Razzaq L, Abbas MM, Miran S, Asghar S, Nawaz S, Soudagar MEM, et al. Response Surface Methodology and Artificial
Neural Networks-Based Yield Optimization of Biodiesel Sourced from Mixture of Palm and Cotton Seed Oil. Sustaina-
bility (Switzerland) 2022;14. https://doi.org/10.3390/su14106130.

[21] Vinoth Arul Raj J, Praveen Kumar R, Vijayakumar B, Gnansounou E, Bharathiraja B. Modelling and process optimiza-
tion for biodiesel production from Nannochloropsis salina using artificial neural network. Bioresour Technol 2021;
329:124872. https://doi.org/10.1016/j.biortech.2021.124872.

[22] Garg A, Jain S. Process parameter optimization of biodiesel production from algal oil by response surface methodology
and artificial neural networks. Fuel 2020; 277:118254. https://doi.org/10.1016/j.fuel.2020.118254.

[23] Ayoola AA, Hymore FK, Omonhinmin CA, Babalola PO, Bolujo EO, Adeyemi GA, et al. Data on artificial neural net-
work and response surface methodology analysis of biodiesel production. Data Brief 2020; 31:105726.
https://doi.org/10.1016/j.dib.2020.105726.
[24] Soji-Adekunle AR, Asere AA, Ishola NB, Oloko-Oba IM, Betiku E. Modelling of synthesis of waste cooking oil methyl
esters by artificial neural network and response surface methodology. International Journal of Ambient Energy 2019;
40:716–25. https://doi.org/10.1080/01430750.2017.1423378.

[25] Muhammad G, Ngatcha ADP, Lv Y, Xiong W, El-Badry YA, Asmatulu E, et al. Enhanced biodiesel production from wet
microalgae biomass optimized via response surface methodology and artificial neural network. Renew Energy 2022;
184:753–64. https://doi.org/10.1016/j.renene.2021.11.091.
[26] Ijaz Malik MA, Usman M, Hayat N, Zubair SWH, Bashir R, Ahmed E. Experimental evaluation of methanol-gasoline
fuel blend on performance, emissions and lubricant oil deterioration in SI engine. Advances in Mechanical Engineering
2021; 13:16878140211025212.

[27] Alhassan FH, Rashid U, Taufiq-Yap YH. Optimization of simultaneous production of waste cooking oil based-biodiesel
using iron-manganese doped zirconia-supported molybdenum oxide nanoparticles catalyst. Journal of Renewable and
Sustainable Energy 2016;8. https://doi.org/10.1063/1.4953090.

[28] Yusuff AS, Popoola LT, Adeniyi DO, Olutoye MA. Coal fly ash supported ZnO catalyzed transesterification of Jatropha
curcas oil: Optimization by response surface methodology. Energy Conversion and Management: X 2022; 16:100302.
https://doi.org/10.1016/j.ecmx.2022.100302.

[29] Gupta S, Patel P, Mondal P. Biofuels production from pine needles via pyrolysis: Process parameters modeling and
optimization through combined RSM and ANN based approach. Fuel 2022; 310:122230.
https://doi.org/10.1016/j.fuel.2021.122230.

[30] Ahmad G, Imran S, Farooq M, Shah AN, Anwar Z, Rehman AU, et al. Biodiesel Production from Waste Cooking Oil
Using Extracted Catalyst from Plantain Banana Stem via RSM and ANN Optimization for Sustainable Development.
Sustainability (Switzerland) 2023;15. https://doi.org/10.3390/su151813599.