Reaction Kinetics, Mechanisms and Catalysis

https://doi.org/10.1007/s11144-023-02360-9

Intervention of artificial intelligence to predict

the degradation and mineralization of amoxicillin
through photocatalytic route using nickel
phosphide‑titanium dioxide catalyst

Sheetal Sethi1 · Amit Dhir1 · Vinay Arora2

Received: 2 January 2023 / Accepted: 21 January 2023

© Akadémiai Kiadó, Budapest, Hungary 2023

Abstract
This research work aims to assess the efficacy of the lab synthesized catalyst ­Ni2P–
TiO2 (NPT) using Artificial neural network (ANN) for the degradation of Amoxicil-
lin (AMX) in aqueous suspension under UV irradiation. The experiments were con-
ducted at 50 ppm antibiotic concentration, using three different compositions of the
synthesized catalyst (1:9, 3:7, 5:5) for 5 h. Of the various catalysts tested, the opti-
mum pH conditions, dose, and time were attained i.e., natural pH, 0.25 g/L, 2 h. The
degradation and mineralization emerged highest with the respective percentages of
83.00 and 70.00%. ANN was applied with the Swish activation Function to predict
amoxicillin degradation. Chemical oxygen demand (COD) removal was considered
the key parameter for determining amoxicillin degradation using a three-layer back-
propagation neural network. The results obtained through the ANN were similar
to the experimental results, and their correlation coefficient was 0.96. The findings
show that all the input variables such as pH, catalyst dose, and irradiation time have
an immense effect on the degradation efficiency. The study demonstrates that Neural
Network modeling can successfully predict and simulate the degradation process.

Keywords  Photocatalysis · Degradation · Machine learning · Artificial neural

network · Backpropagation neural network · Swish activation function

* Amit Dhir
[email protected]
1
School of Energy and Environment, Thapar Institute of Engineering and Technology, Patiala,
India
2
Computer Science & Engineering Department, Thapar Institute of Engineering and Technology,
Patiala, India

13
Vol.:(0123456789)
 Reaction Kinetics, Mechanisms and Catalysis

Introduction

Because of the general lack of resistance to disease among humans, antibiotic

use is always precarious. These medications enter the environment via sewage,
which contains un-metabolized pharmaceuticals from humans and hospital and
industrial waste [1, 2]. In addition to being oxidized by a sophisticated oxidation
mechanism, AMX has the distinction of being the most long-lasting antibiotic.
The antibiotic amoxicillin is commonly recommended since it is a member of
the penicillin class of drugs. It is made up of beta-lactam rings, and even trace
amounts can promote antibiotic-resistant bacteria to flourish [3].There are a num-
ber of reports in the scientific literature stating that pharmaceutical chemicals
and their metabolites are not entirely removed by standard wastewater treatment
plants and needs specialized treatments [4].Consequently, several techniques,
including the improved oxidation process, adsorption, coagulation, flocculation,
reverse osmosis, and other separations methods, have been explored to detoxify
antibiotics from aqueous streams [5, 6].
Advanced oxidation processes (AOPs) are considered useful for degrading var-
ious harmful diligent organic compounds into end products such as carbon diox-
ide and water [7, 8]. The photocatalytic reaction entails heterogeneous catalysis
wherein a semi-conductor is used to absorb the photon to degrade the pollutants
that adhere to the catalyst’s surface [9]. During the photocatalysis process, the
harmless by-products produced as the process depend upon the release of elec-
trons and the formation of a positive hole [11]. Several experimental parameters
evaluate the performance of AOP, such as temperature, light intensity, catalyst
dose, pH, etc., on the degradation process [10]. Each variable plays a significant
role in the degradation process; experimental conditions must optimize for better
results [12].
In order to reduce experimental error and time, various modelling techniques
have been considered to predict and optimize process parameters through AI
models. Algorithms that can learn, recognize patterns, improve accuracy, and
remain stable are employed in many areas of engineering. Zhang and Xu made
predictions on how certain physical characteristics, such as particle size, volume
fraction, thermal conductivity, etc., will affect the nanofluids’ thermal perfor-
mance. To determine the effect of various physical factors on thermal conductiv-
ity improvement, the authors employ a Gaussian process regression model with
99.99 percent correlation coefficient and a 0.0017% root mean square error [13].
The mechanical parameters of a wheat straw reinforced polypropylene compos-
ite were estimated using a Gaussian process regression model, and the results
showed a high degree of accuracy [14]. Because of its affordability, dependabil-
ity, and accuracy, the least square boosting model was employed by the authors to
forecast the modulus elasticity and molar solubility of supercritical carbon diox-
ide in 24 different widely used ionic liquids. [15, 16]. To determine the statistical
­ gB2 superconductors at room tem-
connection between the critical resistivity of M
perature, the authors employ a regression trees model. The tree-based regression
analysis correctly captured MgB2’s disorder and defining features [17].

13
Reaction Kinetics, Mechanisms and Catalysis

There are various others commonly used artificial intelligence (AI) techniques
simulate the different decontamination strategies. The most commonly used algo-
rithms in previous reported researches are ANN, support vector machine (SVM)
and response surface methodology (RSM), genetic algorithm. An artificial neural
network algorithm is a non-linear, easily trainable, adaptive structure, and train-
ing parameters can be tuned. ANN works on the mechanism of the biological type
of neural system by mimicking the structure and function of the human brain. In
this algorithm, the weighted sum of the inputs reaching each neuron passes over
an activation function through which an output signal is generated [18]. Different
types of ANNs have been used to study, like feed-forward (FF) backpropagation,
cascade-forward backpropagation (CF), and the Elman backpropagation network
[19].
The ANN model can be developed with little to no specialised knowledge of
statistics because of its simplicity, ease, and accessibility [20]. In a very short
amount of time, it is possible to identify intricate nonlinear connections between
independent and dependent variables [21]. ANN’s capacity to optimise operating
parameters to save energy consumption has increased its popularity for applica-
tion in the wastewater treatment sector [22, 23].
Jha et al. used ANN and RSM to find the COD removal and hydrogen yield,
and the results show that 84.41% removal efficiency was attained with a 1.01%
error [24]. Counterfeit ANN can be considered the best elective for optimizing
and predicting the process. Bararpour et al. implemented RSM and ANN and
revealed that ANN could anticipate more successfully than RSM with 96% effi-
ciency after a 45-min introduction of light, which is quite close to actual experi-
mental values [25]. Chesterfield and Adesina developed ANN to model the pho-
todegradation process as a guide in designing a new photocatalyst [26]. Zarai et
al. used ANN to predict the colour removal of the BR46 solution; it predicted an
outstanding performance with 98% ­R2 value [27]. The use of ANNs and support
vector machines (SVMs) was assessed for demand forecasting with a colour com-
ponent in the retail clothing business was investigated and ANN was reported to
be better performing tool when compared to SVMs in terms of root mean square
error (RMSE) [28].
The present study aims at photocatalytic decomposition of an Amoxicil-
lin compound in aqueous media using a lab synthesized catalyst i.e. N ­ i2P–TiO2
(NPT) and process optimization was attained through Swish activation function
used in ANNs. The authors here investigated how several invitro characteris-
tics, such as pH, catalyst dose, and irradiation period, affected the degradation of
Amoxicillin measured in terms of COD removal. Generally, there is widespread
application of the Sigmoid function as an activation function in ANNs which has
the limitations of being more prone to issues during training due to its vanishing
gradients, hence the authors have implemented an alternative Swish activation
function because of its unboundedness quality, which allows it to avoid training
delays caused by harsh zero gradients. The findings will be helpful in efficient
treatment of wastewater in treatment plants under optimized dosages of catalyst
and irradiation periods in conjunction with existing conventional treatments.

13
 Reaction Kinetics, Mechanisms and Catalysis

Experimental section

Materials and methods

Amoxicillin was purchased from Sigma Aldrich Company (India) and used as
received. Nickel chloride (hexahydrate), titanium dioxide, and sodium hypophos-
phite were used for the catalyst synthesis. These were arranged by Loba Chemie Pvt.
Ltd. India. All the experiments were executed with distilled water and a batch-type
laboratory-scale photo-reactor with 7 UV tubes of 36W each with a wavelength of
365 nm as UV radiation sources. The stock solution of AMX in desired concentra-
tion was prepared for a typical photocatalytic run.

Synthesis of catalyst

The catalyst was prepared through the wet impregnation method using ­NiCl2·6H2O
and ­TiO2. All reagents were used for the process in their original form analytical
grade. ­NiCl2·6H2O (0.2–1  g) and ­TiO2 (1–1.8  g) were dispersed in distilled water
under constant magnetic stirring for 5 h, then placed overnight for drying in an oven
at 80 °C. N
­ i2P–TiO2 was prepared by annealing with 500 mg sodium hypophosphite
at 300 °C for 2 h. Then, the resultant powder was washed three times with deionized
water. The prepared catalyst was placed in an oven at 80 °C and calcined at 300 °C
in a Muffle furnace. Almost 2 g of catalyst synthesized; three different composites of
NPT having (1:9, 3:7, 5:5) compositions prepared.

Synthesized catalyst characterization

X-ray diffraction (XRD) determined the crystalline phases and composition of

the synthesized catalyst (XRD, Cu ­Kα radiation 1.5406  Å, 45  kV 40  mA) with a
PW3050/60 Panalytical Xpert-PRO diffractometer was employed for the analysis
of the phase Composition and quantitative structure of the lab synthesized cata-
lyst was performed using Rietveld refinement by High score plus. Scanning Elec-
tron Microscopy Dispersive X-ray spectroscopy was also done to study morphol-
ogy and chemical composition. Scanning electron microscopy (SEM) based on the
interaction of the material with a focused electron beam with energies of between
0.5 and 35 kV was used to measure the surface morphology of solid materials. The
Brunauer–Emmett–Teller (BET) surface area was measured by a Quantachrome
sorption analyzer based on N­ 2 adsorption–desorption isotherms.

Photoreactor description

A Photoreactor is used to conduct the Photocatalysis process in a lab-scale designed

photoreactor consisting of 7 UV tubes arranged parallel on the top of the roof. These
UV tubes were used as an irradiation source, emitting UV radiation predominantly at

13
Reaction Kinetics, Mechanisms and Catalysis

365 nm. The sample was stirred continuously using a magnetic stirrer throughout this
photoreactor’s reaction. An exhaust fan has been fitted for ventilation and removal of
the excess heat.

Experimental procedure

All the experiments were executed in a batch-type manner using a laboratory scale pho-
toreactor with the 50 ppm amoxicillin solution in a 250 mL vessel. For each reaction,
100 ml of amoxicillin solution was taken; HCl and NaOH were used to adjust the pH.
The prepared catalyst was added to the solution for the degradation studies. Among the
various parameters, pH (3–9), catalyst dose (0.25–0.50 g/L), and irradiation exposure
(5 h) were selected to measure the activity of the synthesized catalyst for the degra-
dation of amoxicillin under UV radiation. After every hour aliquots were withdrawn
from the reactor for liquid analysis in terms of COD removal. The sample was filtered
through a hydrophilic PTFE Miller LCR filter having a pore size of 0.22 µm. All exper-
iments were performed thrice to assess the accuracy of the results.

Analytical techniques

COD is determined by the closed reflux method. The appropriate samples were pre-
pared by adding dichromate, sulfuric acid, and mercuric sulfate and then kept for diges-
tion in a COD digester for 2  h. The sample was titrated against ferrous ammonium
sulphate. At the optimized condition, Total organic carbon was also measured using
the Total Organic Carbon (TOC) analyzer by Multi N/C 2100 BU (Analytik Jena AG
Corporation, Germany). It is based on the phenomenon of catalytic combustion/ non-
dispersive infrared gas analysis. AMX degradation and mineralization efficiency are
calculated by the following equation [29]:
( )
[ ] [AMX]i − [AMX]F
Degradation Efficiency % = × 100, (1)
[AMX]i

( )
[ ] [TOC]i − [TOC]F
Mineralization Efficiency % = × 100, (2)
[TOC]i

[AMX]i Initial AMX concentration (mg/L), [AMX]F Final AMX concentration after
treatment (mg/L), [TOC]i Initial total organic carbon concentration (mg/L),[TOC]F
Total organic carbon concentration after the photocatalytic treatment (mg/L).

Results and discussions

The catalyst was synthesized in the laboratory through the wet impregnation method
by adding the appropriate amount of nickel chloride and titanium dioxide. The three
different compositions (1:9, 3:7, 5:5) of nickel phosphide-titanium dioxide were

13
 Reaction Kinetics, Mechanisms and Catalysis

prepared. The stock solution of the AMX compound was prepared for the deg-
radation studies, and the 50  ppm AMX solution was assessed for its degradation
efficiency.

Characterization of NPT

To get an insight into the synthesized catalysts, the characterization was done. In
this study, X-Ray diffraction, SEM–EDX, and BET, of the synthesized catalyst were
used to characterize the catalyst.
Fig.  1 highlights the samples’ XRD pattern to show the structure’s nature. The
diffraction peaks of the sample (1:9) at 30.48°, 31.80°, 40.72°, 44.60°, 47.31°,
54.25°, 54.96°, 63.49, 66.28°, 66.39, 72.63, 74.77 were well-matched with the hex-
agonal phase structure of Nickel Phosphide (NP), which corresponded to (110),
(011), (111), (021), (120), (002), (121), (112), (130), (022), (131), and (122) lat-
tice planes of the standard card of NP(JCPDS#65-3544). Expert high score plus was
used to analyse the phase compositions of the synthesized catalysts. The measure-
ment of d-spacing (0.28  nm) was in good agreement with XRD patterns, as men-
tioned earlier [30, 31].
Scanning electron microscopy (SEM) forms images explaining the material’s
various properties, e.g., topography, elemental composition, and composition het-
erogeneities. Fig.  2 displays the microstructure of the catalyst ­Ni2P–TiO2. The
integration of N­ i2P with T­ iO2 was expected to enhance the photocatalytic activ-
ity, and it was seen that N­ i2P grew uniformly. The elemental mapping results Ni
and Ti coexisted. To get an insight into the catalyst, an EDS analysis was done to

6000

Ni2P- TiO2(5:5)
Intensity (a.u.)

4000 Ni2P- TiO2(3:7)

Ni2P- TiO2(1:9)

2000

0
20 30 40 50 60 70 80

2θ (°)

Fig. 1  The XRD Patterns Obtained for three Different Compositions of NPT (1:9, 3:7, 5:5)

13
Reaction Kinetics, Mechanisms and Catalysis

(a) (b)

200 nm 200 nm

(c) (d)

400 nm 2µm

(e)

Fig. 2  SEM images of lab synthesized catalysts a 1:9 NPT, b, c 2:8 NPT, d 3:7 NPT, e EDX Image of
(1:9) NPT Composite

quantify the content of the photo-catalyst having the composition (1:9). The EDS
of (1:9) showed the presence of carbon, oxygen, sodium, titanium, and nickel
with the compositions as C—3.21, O–47.21, Na—1.17, Ti—44.54, Ni—0.93%.

13
 Reaction Kinetics, Mechanisms and Catalysis

BET of the lab synthesized catalysts was performed to confirm the surface area of
the catalyst. The synthesized catalyst (1:9), which constituted only one ratio of nickel
phosphide and the remaining titanium dioxide, had a surface area of 10.758m2/g.

Catalytic activity of the NPT

The catalyst’s different compositions have been assessed to find the degradation effi-
ciency of synthesized catalysts. In this current study, the effect of catalyst loading,
pH, and irradiation exposure has been evaluated to assess the degradation efficiency
of amoxicillin.

Effect of pH

Hydroxyl ion generation is strongly dependent on pH. Experiments were performed

with different pH values ranging from 3 to 9 to study the effect of pH photocatalytic
degradation of amoxicillin (50 ppm). The solution mixed with the different doses of
the prepared photocatalyst ranging from 0.25 to 0.50 g/L was placed in the photore-
actor under constant and continuous UV light. The results showed that 83.0% degra-
dation in 2 h at natural pH in terms of COD removal (%) with 0.25 g/L catalyst dose.
Among the three different compositions of NPT (1:9, 3:7, 5:5) and ­TiO2, the compo-
sitions in the (1:9) ratio showed maximum degradation. Amoxicillin has three pKa
values at 2.7, 7.5, and 9.6, whereas AMX is undoubtedly positively charged at pH 5,
so is the surface of Degussa P25, the point of zero charge, 6.7. The particles of ­TiO2
started to agglomerate when dispersed in aqueous media; agglomeration of particles
in water diminished beneath acidic conditions driven to the effective surface area of
the catalyst [32, 33].

Effect of the catalyst dose

Catalyst dose was varied in the range of 0.25–0.50  g/L to observe the efficacy of
synthesized catalyst on the degradation of AMX. Fig.  3 indicates the degradation
of AMX in 120  min at a natural pH and 0.25–0.50  g/L of catalyst dose. The four
catalysts (1:9) showed 83% degradation with 0.25 g/L in 2 h, whereas no significant
antibiotic degradation was observed on increasing the concentration of the catalyst
over 0.25 g/L. It may be attributed to the turbidity of the suspension; if the suspen-
sion gets turbid at that point, UV light infiltration diminishes, and restricted illumi-
nation reaches the suspension. The results achieved were in good agreement with
the other reported literature. Verma and Haritash reported that after a specific cata-
lyst dosage, degradation of the compound reduced due to lack of light penetration
and agglomeration of the ­TiO2 catalyst [34]. On increasing the concentration of the
catalyst, the screening effect of the overabundance of the particles happened so that
it prevented light entrance to quite a large extent [4, 35]. Sierra et al. reported the
clouding effect as catalyst dose increases; photons are not allowed to pass through
suspension, thus reducing the formation of electron–hole pair [36].

13
Reaction Kinetics, Mechanisms and Catalysis

(A)
(B)

(C)

Fig. 3  Catalytic activity of the lab synthesized catalyst. A Effect of various pH on AMX degradation in
terms of COD removal (%) using T ­ iO2 and different composition of NPT (1:9, 3:7, 5:5) at (after 120 min
and with 0.25  g/L catalyst dose, 83% degradation achieved with (1:9) composite at natural pH(8), B
Effect of Irradiation time (60–300 min) at normal pH, 0.25 g/L catalyst dose, 83% degradation achieved
in 120  min using (1:9) NPT Composite, C Effect of catalyst dose (0.25–0.50  g/L), where at 0.25  g/L
optimized dose with 2 h irradiation exposure, (1:9) NPT catalyst

Effect of irradiation time

Time has a significant role in the mass transfer rate of the pollutants during the adsorp-
tion process. The suspension was placed in the photoreactor under constant stirring in
the dark to achieve absorption/desorption equilibrium. UV irradiation exposure was
provided to the solution continuously for 5 h with constant stirring. For the test pon-
ders, ­TiO2 and three different compositions of ­Ni2P–TiO2 (1:9, 3:7, 5:5) were inspected
to find the effective one. ­TiO2 gave 33.33% COD removal in 120  min compared to
­Ni2P–TiO2 (1:9), which gave 83% COD removal in a fair 120 min at 0.25 g/L catalyst
dose at natural pH. COD removal efficiency increased up to 120 min, then decreased
due to diminishing the official destination with expanding the contact time [37]. Wah-
yuni et al. reported the effectiveness of the light and the catalyst; after some time, pho-
todegradation reduced due to exhaustion of the catalyst [38].

Mineralization study

During the Photocatalytic degradation process, compound transformed into ­CO2,

­H2O and inorganic ions. In order to get confirmation, mineralization studies were

13
 Reaction Kinetics, Mechanisms and Catalysis

done to assess total organic carbon (TOC) present in the sample under UV irradia-
tion. Although the degradation of amoxicillin was about 83% at natural pH, but 70%
mineralization occurred in 120 min.

Artificial neural network model development

The ANN model is widely used for classification, clustering, and prediction in vari-
ous fields. It is beneficial in conventional regression and statistical models [39]. The
primary benefit of using ANN is its applicability to find out patterns even if the
data is noisy [40]. The artificial neural network has neurons like the human brain,
where a neuron consists of dendrites, nucleus, axons, and synapses. In the human
neural system, dendrites receive input from the other neurons; and axons transmit
from one to another. The ANN resembles the human neural system, consisting of
input, connection weights, summing function, computation, and an output variable
[41]. Hidden layers exist in both of them. The degrees of accuracy in these layers
can be checked out, hidden neurons perform as feature detectors, and the number
of neurons may vary. ANN has specific characteristics to implement the non-linear
approximation function [42]. It has been broadly reported that the sigmoid function
is most widely used. Lenzi G.G. used a sigmoid activation function for the degrada-
tion of textile dye and achieved the most promising results with correlation coeffi-
cients of 0.99 and 0.98 [43].
The experimental data were classified into three sets: data to train, test, and vali-
date. Approx. 60% of obtained data was used to train the model. A swish activa-
tion function was employed as the transfer function for the neurons at the input and
output layers. A three-layer backpropagation model was developed to train and test
the data. There are three input variables, viz. pH, catalyst dose, irradiation time, and
one output variable in terms of COD removal. Experimental data has been generated
through synthesized photocatalyst and implementation of the ANN done to stimu-
late and predict the degradation process.
In the previous studies, sigmoid was employed widely as the activation function,
and maximum output was computed with the help of MATLAB [44, 45]. Rasouli-
fard M.H. implemented the sigmoid activation function to check the photocatalytic
activity of chitostan in the presence of visible light.
1
F(X) = (3)
1 + e−x
The model predicted decolorization efficiency in terms of the ­R2 value (0.98)
[46]. The Swish activation function has been used in this report due to its smooth
and bound below, but there is no upper bound [47]. While tuning the model offers a
significant role in the minimal loss.
X
Swish(x) = x × sigmoid(𝛽x) = (4)
1 + e−𝛽x

13
Reaction Kinetics, Mechanisms and Catalysis

Here 𝛽 is either a constant or trainable parameter and the Swish activation function
is used in reinforcement learning [48].
Data was generated by conducting the experiments in the laboratory with the
help of a synthesized catalyst. Three input variables and one output variable were
selected for the implementation of ANN on the generated experimental data through
Photocatalysis (Table 1).
The performance of the network was assessed by statistical calculation correla-
tion in terms of coefficient (R) from Eq. (5), normalized Mean square error (NMSE)
Eq. (7), and Normalized value of Root Mean Square Error (NRMSE) Eq. (6), MSRE
values Eq. (8), and Theil values with Eq. (9) as follows [52–54]:
�
� ∑N � �2
�
� Xpred(i)− Xexp(i)
�
R = �1 −
i=1
(5)
∑N � �2
i=1
X pred(i) − X 1

�
1 ∑n � �2
t − oi
NRMSE =
N i=1 i
(6)
tmax − tmin

∑N � �2
i=1
Xpred(i) − Xexp(i)
NMSE = ∑N � �2 (7)
i=1
Xexp(i)

n (
|1 ∑ )
| ti − oi 2 ||
MSRE = | | (8)
|n ti |
| i=1 |

�
⎡ ∑n � �2 ⎤
⎢ t − oi ⎥
i=1 i
U2theil =⎢ �∑ ⎥ (9)
⎢ n
t 2 ⎥
⎣ i=1 i ⎦

Some errors occurred during the training, testing, and validation at different
steps, but the error was found with the help of statistical evaluations. The NMSE,

Table 1  Advantages of swish activation function over sigmoid activation function

S. No. Sigmoid activation function Swish activation function References

1 Non-linear, Logistic function Hybrid function [49]

2 Expressed as:F(X) = 1 X [47]
1+e−x 1+e−𝛽x
3 Output range (0,1) output range (− ∞, ∞) [50]
4 Gradient vanish, Not symmetric about It can be used in the case of zero [51]
zero derivatives

13
 Reaction Kinetics, Mechanisms and Catalysis

NRMSE, MSRE, R ­ 2, Theil, etc., formed the evaluation criteria. The correlation
coefficient explained the correlation between the predicted and experimental. It
was observed that the value for ­R2 should be approximately one [55].
The performance of the mode was verified with statistical evaluation. As
reflected in Table  2, the NMSE and NRMSE values were 0.0121 and 0.2179,
respectively, with a minimum Theil value of 0.125. It confirmed the robustness of
the model to depict the degradation process with excellent accuracy.

Validation of experiments conducted through swish‑based ANN model

The process followed in the optimized model for amoxicillin degradation has
been depicted. The ANN model was trained, tested, and validated with the help of
the Swish activation function. The experimental results showed significant results
at normal pH, with 0.25 g/L catalyst dose, in 120 min of irradiation. The correla-
tion coefficient value indicated that ANN could efficiently interpret and predict
the degradation process. The ANN results demonstrated that each experimental
run correlated strongly with the R ­ 2 (0.960) correlation coefficient predicted val-
ues. From the results, it can be concluded that ANN utilized Swish as an acti-
vation function applied successfully to simulate the relation between input and
output variables of the degradation process.
Fig.  4 reveals the proximity of predicted values versus experimental COD
removal percent values. The results indicated that the fitness of the deployed
model was able to predict the performance successfully with great accuracy. The
­R2 value of 0.960 obtained from the ANN prediction clearly explained the mod-
el’s suitability to effectively learn the relationship between the input parameters
and output variables.
Fig. 5 described the successful implementation of the model; it also provided
the information transferred to output for the prediction of degradation of the com-
pound with minimal residual. The literature described the effectiveness of the
ANN-based model in different fields. Zulfikar et al., in their work, applied ANN
modelling for depicting degradation efficiency by incorporating different input
variables and described the non-linear relationship among them [56]. Basturk
and Alver displayed that the ANN model could predict the removal with minimal
residual error and provided better results than Response Surface Methodology
(RSM) [57].

Table 2  Results obtained after Tool Evaluation

successfully implementing the
ANN Model R2 0.96
NMSE 0.0121
NRMSE 0.2179
MSRE 0.0384
Theil 0.125

13
Reaction Kinetics, Mechanisms and Catalysis

Fig. 4  Experimental Versus Predicted Values for AMX Degradation in terms of COD removal (%) by
employing the ANN model using swish activation function. To train the model, three input parameters
(pH, catalyst dose, irradiation time) has been taken and COD removal (%) as an output

100 y = 0.6736x + 16.453

R² = 0.9604
Predicted COD Removal

80

60
(%)

40

20

0
0 10 20 30 40 50 60 70 80
Actual COD Removal (%)

Fig. 5  Experimental COD Removal (%) versus Predicted COD Removal (%) for the deployed model with
­R2 = 0.96

Conclusion

The present study focused on the degradation of AMX through a photocatalytic pro-
cess via lab synthesized catalyst, and the parametric optimization was done using
ANN. Results revealed that at Natural pH with 0.25 g/L catalyst dose and 2 h of irra-
diation, 83.00% degradation of amoxicillin antibiotic was attained in terms of COD
removal. The 70.0% mineralization of the compound was also confirmed through
a TOC analyzer under optimized conditions. A three-layer backpropagation neu-
ral network was optimized to degrade Amoxicillin in aqueous media. In this neural

13
 Reaction Kinetics, Mechanisms and Catalysis

network, the Swish activation function was employed and the findings showed ANN
predicted outcomes with a correlation coefficient of 0.96. Hence, the problem was
resolved efficiently through ANN to discover the most suitable working circum-
stance for the efficient photocatalytic degradation. The acquisition of original data
for model training is relatively costly, making the training of ANN another difficulty
in time series databases for experimental investigations. The current study used a
synthetic antibiotic solution, but in real wastewater, the concentration of the anti-
biotic may be different and there is a possibility of several other chemical interfer-
ences. Moreover, the study was conducted under UV irradiations and in order to
optimize the cost, further studies needs to be conducted using solar light induced
degradation in real wastewater containing such antibiotics.

Authors contributions  All authors contributed to the study’s conception and design, where SS performed
the material preparation, experimentation, analysis, and draft preparation. All authors (AD and VA) com-
mented on the previous version of the manuscript. All authors have read and agreed to the published ver-
sion of the manuscript.

Funding  The authors declare that no funds, grants, or other support were received during the preparation
of this manuscript.

Data availability  We guarantee the authenticity and reliability of the data.

Declarations 
Competing interests  The authors declare they have no financial interest.

Ethical approval  Not Applicable.

Consent to participate  Not Applicable.

Consent to publish  All authors have known and agreed to publish this article in the Journal of environ-
mental science and pollution research.

References
1. Ellepola N, Rubasinghege G (2022) Heterogeneous photocatalysis of amoxicillin under natural
conditions and high-intensity light: fate, transformation, and mineralogical impacts. Environments
9(7):77
2. Karamanlis VDLXN, Koveos SPPDS (2018) Effects of the antibiotic amoxicillin on key species of
the terrestrial environment. Bull Environ Contam Toxicol 100:509–515
3. Ayodele OB, Auta HS, Nor N (2012) Artificial Neural networks, optimization and kinetic modeling
of amoxicillin degradation in photo-fenton process using aluminum pillared montmorillonite-sup-
ported ferrioxalate catalyst. Ind Eng Chem Res 51:16311–16319
4. Dimitrakopoulou D, Rethemiotaki I, Frontistis Z et al (2012) Degradation, mineralization and anti-
biotic inactivation of amoxicillin by UV-A/TiO 2 photocatalysis. J Environ Manage 98:168–174
5. Champdore MDE, Zuccato E (2004) Antibiotics in the environment: occurrence in Italian STPs,
fate, and preliminary assessment on algal toxicity of amoxicillin. Environ Sci Technol 38:6832–6838
6. Homem V, Santos L (2011) Degradation and removal methods of antibiotics from aqueous matri-
ces—A review. J Environ Manage 92:2304–2347
7. Mohammed S, Fasnabi PA (2016) Removal of dicofol from waste-water using advanced oxidation
process. Procedia Technol 24:645–653

13
Reaction Kinetics, Mechanisms and Catalysis

8. Andreozzi R, Caprio V, Insola A, Marotta R (1999) Advanced oxidation processes (AOP) for
water purification and recovery. Catal Today 53:51–59
9. Ra A, Ikram M, Ali S et al (2021) Photocatalytic degradation of dyes using semiconductor pho-
tocatalysts to clean industrial water pollution. J Ind Eng Chem 97:111–128
10. Wang J, Wang S (2021) Effect of inorganic anions on the performance of advanced oxidation
processes for degradation of organic contaminants. Chem Eng J 411:128392
11. Ali MHH, Al-qahtani KM, El-sayed SM (2019) Enhancing photodegradation of 2, 4, 6 trichlo-
rophenol and organic pollutants in industrial effluents using nanocomposite of TiO 2 doped with
reduced graphene oxide. Egypt J Aquat Res 45:321–328
12. Homem V, Alves A, Santos L (2010) Amoxicillin degradation at ppb levels by Fenton’s oxida-
tion using design of experiments. Sci Total Environ 408:6272–6280
13. Zhang Y, Xu X (2020) Predicting the thermal conductivity enhancement of nanofluids using
computational intelligence. Phys Lett A 384:126500
14. Zhang Y, Xu X (2021) Machine learning with applications machine learning tensile strength and
impact toughness of wheat straw reinforced composites. Mach Learn with Appl 6:100188
15. Zhang Y, Xu X (2022) Modulus of elasticity predictions through LSBoost for concrete of normal
and high strength. Mater Chem Phys 283:126007
16. Zhang Y, Xu X (2020) Solubility predictions through LSBoost for supercritical carbon dioxide in
ionic liquids. New J Chem 44:20544–20567
17. Zhang Y, Xu X (2022) Superconductivity and its applications Disordered MgB 2 superconductor
critical temperature modeling through regression trees. Phys C: Supercond Appl 597:1354062
18. Al-Araimi MM, Varghese MJ, Nageswara Rao LS, Feroz S (2019) Optimization and assessment
of residual chlorine using response surface methodology (RSM) and artificial neural network
(ANN) modeling. Int J Recent Technol Eng 8:258–263
19. Fidalgo A, Letichevsky S, Santos BF (2021) Assessment of T ­ iO2 band gap from structural
parameters using artificial neural networks. J Photochem Photobiol A Chem 405:112870
20. Tu JV (1996) Advantages and disadvantages of using artificial neural networks versus logistic
regression for predicting medical outcomes. J Clin Epidemiol 49:1225–1231
21. Narendra KS (1996) Neural networks for control: theory and practice. Proc IEEE 84:1385–1406
22. Frontistis Z, Daskalaki VM, Hapeshi E et al (2012) Photocatalytic (UV-A/TiO 2) degradation of
17α- ethynylestradiol in environmental matrices: experimental studies and artificial neural net-
work modeling. J Photochem Photobiol A Chem 240:33–41
23. Dutta S, Parsons SA, Bhattacharjee C et al (2010) Development of an artificial neural network
model for adsorption and photocatalysis of reactive dye on TiO2 surface. Expert Syst Appl
37:8634–8638
24. Jha P, Kana EBG, Schmidt S (2017) Can artificial neural network and response surface meth-
odology reliably predict hydrogen production and COD removal in an UASB bioreactor ? Int J
Hydrogen Energy 42:18875–18883
25. Bararpour ST, Feylizadeh MR, Delparish A et al (2018) Investigation of 2-nitrophenol solar deg-
radation in the simultaneous presence of K2S2O8 and H2O2: using experimental design and
artificial neural network. J Clean Prod 176:1154–1162
26. Chesterfield D, Adesina AA (2009) Evidence-based design and optimisation of titania photocata-
lysts via artificial neural network analysis. J Chem Eng Japan 42:185–198
27. Zarei M, Khataee AR, Ordikhani-Seyedlar R, Fathinia M (2010) Photoelectro-Fenton combined
with photocatalytic process for degradation of an azo dye using supported TiO2 nanoparticles
and carbon nanotube cathode: neural network modeling. Electrochim Acta 55:7259–7265
28. Güven İ, Şimşir F (2020) Demand forecasting with color parameter in retail apparel industry
using artificial neural networks (ANN) and support vector machines (SVM) methods. Comput
Ind Eng 147:106678
29. Souza FS, Vargas V, Rosin CK et  al (2017) Comparison of different advanced oxidation pro-
cesses for the removal of amoxicillin in aqueous solution. Environ Technol 39:549–557
30. Zhang Y, Wang G, Jin Z (2019) An orderly assembled g-C 3 N 4, rGO and Ni 2 P photocatalyst
for efficient hydrogen evolution. Int J Hydrogen Energy 44:10316–10327
31. Liu E, Qi L, Chen J et al (2019) In situ fabrication of a 2D Ni 2 P/red phosphorus heterojunction
for efficient photocatalytic H 2 evolution. Mater Res Bull 115:27–36
32. Elmolla ES, Chaudhuri M (2010) Degradation of amoxicillin, ampicillin and cloxacillin antibiot-
ics in aqueous solution by the UV/ZnO photocatalytic process. J Hazard Mater 173:445–449

13
 Reaction Kinetics, Mechanisms and Catalysis

33. Al-Hamdi AM, Rinner U, Sillanpää M (2017) Tin dioxide as a photocatalyst for water treatment:
a review. Process Saf Environ Prot 107:190–205
34. Verma M, Haritash AK (2020) Photocatalytic degradation of Amoxicillin in pharmaceutical
wastewater : a potential tool to manage residual antibiotics. Environ Technol Innov 20:101072
35. Norabadi E, Hossein A, Ghanbari R, Meshkinian A (2020) Optimizing the parameters of amoxi-
cillin removal in a photocatalysis / ozonation process using Box – Behnken response surface
methodology. Desalin Water Treat 192:234–240
36. Arce-sarria A, Machuca-mart F, Bustillo-lecompte C (2018) Degradation and loss of antibacte-
rial activity of commercial amoxicillin with T ­ iO2/WO3-assisted solar photocatalysis. Catalysts
8:222
37. Liu X, Ma R, Wang X et al (2019) Graphene oxide-based materials for efficient removal of heavy
metal ions from aqueous solution : a review. Environ Pollut 252:62–73
38. Wahyuni ET, Yulikayani PY, Aprilita NH (2020) Enhancement of visible-light photocatalytic
activity of Cu-doped TiO2 for photodegradation of amoxicillin in water. J Mater Environ Sci
11:670–683
39. Abiodun OI, Jantan A, Omolara AE et  al (2018) State-of-the-art in artificial neural network
applications: a survey. Heliyon 4:e00938
40. Beresford R (2000) Basic concepts of artificial neural network ( ANN ) modeling and its applica-
tion in pharmaceutical research. J Pharm Biomed Anal 22:717–727
41. Harvey S, Harvey R (1998) An introduction to artificial intelligence. Appita J. https://​doi.​org/​10.​
2514/6.​1994-​294
42. Ardabili SF, Najafi B, Shamshirband S et  al (2018) Computational intelligence approach for
modeling hydrogen production : a review. Eng Appl Comput Fluid Mech 12:438–458
43. Lenzi GG, Evangelista RF, Duarte ER et al (2016) Photocatalytic degradation of textile reactive
dye using artificial neural network modeling approach. Desalin Water Treat 57:14132–14144
44. Ayodele BV (2020) Backpropagation neural networks modelling of photocatalytic degradation of
organic pollutants using TiO 2 -based photocatalysts. J Chem Technol Biotechnol 95:2739–2749
45. Sahoo C, Gupta AK (2012) Optimization of photocatalytic degradation of methyl blue using
silver ion doped titanium dioxide by combination of experimental design and response surface
approach. J Hazard Mater 215–216:302–310
46. Rasoulifard MH, Dorraji MSS, Amani-ghadim AR, Keshavarz-babaeinezhad N (2016) Applied
catalysis a : general visible-light photocatalytic activity of chitosan / polyaniline / cds nanocom-
posite : kinetic studies and artificial neural network modeling. Appl Catal A GEN 514:60–70
47. Rasamoelina AD, Adjailia F, Sincak P (2020) A Review of Activation Function for Artificial
Neural Network. SAMI 2020 - IEEE 18th World Symposium on Applied Machine Intelligence
and Informatics : proceedings, 281–286. https://​doi.​org/​10.​1109/​SAMI4​8414.​2020.​91087​17
48. Arora V, Mahla SK, Leekha RS et  al (2021) Intervention of artificial neural network with an
improved activation function to predict the performance and emission characteristics of a biogas
powered dual fuel engine. Electron 10:584
49. Mercioni MA, Holban S (2020) P-Swish : Activation Function with Learnable Parameters Based
on Swish Activation Function in Deep Learning. Internation Ssymposium on Electronics and
Telecommunications, IEEE 1–4
50. Pwasong A, Sathasivam S (2016) A new hybrid quadratic regression and cascade forward back-
propagation neural network. Neurocomputing 182:197–209
51. Sharma S, Sharma S (2020) Activation functions in neural networks. Int J Eng Appl Sci Technol
4:310–316
52. Tabatabai-Yazdi FS, Ebrahimian Pirbazari A, Esmaeili Khalil Saraei F, Gilani N (2021) Con-
struction of graphene based photocatalysts for photocatalytic degradation of organic pollutant
and modeling using artificial intelligence techniques. Phys B Condens Matter 608:412869
53. Benramdane IK, Nasrallah N, Amrane A et al (2021) Optimization of the artificial neuronal net-
work for the degradation and mineralization of amoxicillin photoinduced by the complex fer-
rioxalate with a gradual and progressive approach of the ligand. J Photochem Photobiol A Chem
406:112982
54. Pareek VK, Brungs MP, Adesina AA, Sharma R (2002) Artificial neural network modeling of a
multiphase photodegradation system. J Photochem Photobiol A Chem 149:139–146
55. Karaci A, Caglar A, Aydinli B, Pekol S (2016) The pyrolysis process verification of hydrogen
rich gas ( H e rG ) production by artificial neural network ( ANN ). Int J Hydrogen Energy
41:4570–4578

13
Reaction Kinetics, Mechanisms and Catalysis

56. Zulfiqar M, Samsudin MFR, Sufian S (2019) Modelling and optimization of photocatalytic deg-
radation of phenol via TiO2 nanoparticles: an insight into response surface methodology and
artificial neural network. J Photochem Photobiol A Chem 384:112039
57. Baştürk E, Alver A (2019) Modeling azo dye removal by sono-fenton processes using response
surface methodology and artificial neural network approaches. J Environ Manage 248:109300

Publisher’s Note  Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under
a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted
manuscript version of this article is solely governed by the terms of such publishing agreement and
applicable law.

13