Chemometrics and Intelligent Laboratory Systems 242 (2023) 104983

Contents lists available at ScienceDirect

Chemometrics and Intelligent Laboratory Systems

journal homepage: [Link]/locate/chemometrics

Determination of bandgap of period 3, 4, and 5 transition metal dopants on

zinc oxide using an artificial neural network based approach
Muhammad Haris Saeed a, Naveen Kosar a, Sadaf-ul Hassan b, Sohail Nadeem a,
Mazin Abed Mohammed c, *, Mohd Khanapi Abd Ghani d, Karrar Hameed Abdulkareem e
a
Department of Chemistry, University of Management and Technology, Lahore, 54000, Pakistan
b
Department of Chemistry, COMSATS University Islamabad, Lahore Campus, Defense Road, Off Raiwind Road, Lahore, 54000, Pakistan
c
College of Computer Science and Information Technology, University of Anbar, Anbar, 31001, Iraq
d
Biomedical Computing and Engineering Technologies (BIOCORE) Applied Research Group, Faculty of Information and Communication Technology, Universiti Teknikal
Malaysia Melaka, Durian Tunggal, 76100, Malaysia
e
College of Agriculture, Al-Muthanna University, Samawah, 66001, Iraq

A R T I C L E I N F O A B S T R A C T

Keywords: Artificial intelligence (AI) and machine learning (ML) have rapidly emerged as valuable tools for chemical
Photocatalysis research, offering new ways to analyze and understand complex chemical systems. This research article in­
Zinc oxide vestigates the use of adaptive neuro-fuzzy inference system (ANFIS) and multi-layer perceptron (MLP) models to
Predictive analysis
predict the bandgap of transition metal doped zinc oxide (ZnO). The opto-electronic properties of transition
Machine learning
Neural networks
metal doped ZnO complexes are of significant interest because of their applications is optoelectronic systems. The
ANFIS MLP and ANFIS models were trained using a dataset of experimentally measured bandgap values and the cor­
responding structural parameters of the doped ZnO systems. The performance of the models was evaluated using
statistical metrics i.e., RMSE, R, and MAE. The results showed that both MLP and ANFIS models were capable of
accurately predicting the bandgap of transition metal doped ZnO. However, the ANFIS model demonstrated
superior performance with higher accuracy and better generalization ability. The study provides a useful
approach for predicting the bandgap of transition metal doped ZnO using machine learning techniques and may
contribute to the development of advanced optoelectronic devices.

1. Introduction reduced productivity, conflicts over resources, and migration in search

of water [3].
Over the course of time, the use of photocatalysts and sunlight-driven Now-a-days, extensive research is being conducted in the field of
photocatalytic mechanisms are believed to be among the best ap­ semiconductor photocatalysts because of their efficient solar energy
proaches in alleviating prevailing environmental and energy issues [1]. conversion and use in environmental purification [4]. Photodegradation
These materials have gained a significant reputation among scholars as of dyes has been performed under solar and UV radiation. Photocatalysis
they are also projected to be the key step in producing a hydrogen on principle involves the irradiation of photocatalyst with UV or visible
economy. This approach is scientifically supported and the most energy light creating holes because of the absorption of the energy. If the energy
efficient as it directs the focus on the utilization of ~42% sunlight absorbed is more than that of the band gap of the photocatalyst, elec­
(visible region of light) to use and not go to waste. These chemical agents trons jump to the conduction band. In higher energy conduction bands,
have been around for a while and have excellent potential in solving superoxide anions are generated as a result of reaction between dis­
industrial scale dye degradation, and disinfection problems [2]. Water solved oxygen and the electrons while production of hydroxyl radicals is
scarcity can have severe impacts on human lives, leading to a range of due to reaction of holes with water. The production of reactive oxygen
health problems, as well as economic and social challenges such as species allows for the oxidation of harmful organic dyes present in the

* Corresponding author.
E-mail addresses: abdullahsaeed74@[Link] (M.H. Saeed), [Link]@[Link] (N. Kosar), sadafulhassan@[Link] (S.-u. Hassan), sohail.
nadeem@[Link] (S. Nadeem), mazinalshujeary@[Link] (M.A. Mohammed), khanapi@[Link] (M.K. Abd Ghani), Khak9784@[Link]
(K.H. Abdulkareem).

[Link]
Received 8 April 2023; Received in revised form 1 September 2023; Accepted 7 September 2023
Available online 15 September 2023
0169-7439/© 2023 Elsevier B.V. All rights reserved.
M.H. Saeed et al. Chemometrics and Intelligent Laboratory Systems 242 (2023) 104983

industrial wastewater. Organic dyes have versatile uses in several in­ one of the crucial measures for determination of the metal’s photo­
dustries that include paper, leather, cosmetics, etc. These industries catalytic properties. Current computational studies involve the use of
discharge their wastewater in water bodies without any physical or Density Functional Theory and other associated methods for estimation
chemical treatment to remove harmful substances (i.e., organic dyes). of bandgap which are both time intensive and computationally
These toxic organic dyes can accumulate in living organisms leading to burdening tasks. The major contributions of the study are as follows:
health and environmental complications [5]. Many approaches have
been taken to eliminate these toxic organic dyes from the wastewater of 1. Data from studies available in the literature was collected and a
industries. These processes include reverse osmosis, chemical precipi­ neural network-based model was designed for the estimation of
tation, ultracentrifuge filtration, photodegradation, etc. [6]. bandgap of ZnO doped with transition metals from Period 3, 4, and 5.
These properties present semiconductor photocatalysts to be a very 2. This allows for development of a precise synthesis and computational
promising solution for degradation of harmful dyes from industrial tool that can predict the properties of a material beforehand based on
wastewater [7]. In the semiconductor industry, doping is the process of the desired property of the material that is to be synthesized.
adding impurities to a semiconductor to change its properties. Doping is 3. This will help a great deal in determination of optical properties
important to the manufacture of electronics, but it is also important to without any sort of experimental lab work.
the manufacture of solar cells and other semiconductors. Doping allows 4. The factors namely Atomic Radius of dopant (AR), Ionic Radius of
semiconductors to be made with smaller amounts of material, which is the dopant (IR), the Atomic Number of the dopant (An), the Grain
important for manufacturing on a large scale. Doping also changes the Size (GS), and the concentration (C) of the dopant are considered for
bandgap of a semiconductor, which affects its electrical properties. the analysis.
The process of photocatalysis involves the absorption of radiation by 5. An Adaptive Neuro Fuzzy Inference System (ANFIS) and Multi-Layer
electrons present in valence band of photocatalyst. If absorbed radia­ Perceptron (MLP) models are designed using the data collected from
tions are of higher energy than band gap it results in excitation of the literature and different models are trained using different number of
electrons to the higher energy band which is known as the conduction variables for prediction to develop versatility and enabling the
band [8,9]. The optoelectronic properties of a photocatalyst are deter­ ability to be as specific when using the prediction models.
mined by the band gap that is the energy required for an electron to
jump from valence to conduction band. The semiconductors having the The contents of the article are organized in the following manner:
band between 1.5 and 3.0 eV are of special interest as they can absorb Section 3 focuses on providing period-wise fine details on the data
the energy when irradiated with light of visible region of the electro­ values collected from literature for the bandgap prediction process.
magnetic spectrum [8]. While Section 5 addresses the performance of these models using sta­
The electrons get excited by absorbing the light energy results in the tistical metrics, Section 4 provides a brief summary of the methodology
holes having positive charge on the valence band and gets oxidized utilized to create the data models. The research findings are concluded
whereas the conduction band carries a negative charge and gets reduced in Section 6, which also makes a suggestion on the paradigm of studies
[10–12]. After the excitation of the electron there may be several cases that might be developed using the concept employed in the article.
such as the electron may come back to the valence band by the release of
the energy [10,13,14]. The electron-hole pair that is generated due to 2. Artificial intelligence, machine learning and chemistry
absorption of light energy combine very quickly which results in the
reduced photocatalytic activity as very less charges will be present on Scientists working in the field of chemistry can benefit from the
the surface for the degradation of the pollutants [11]. strength of machine learning. To find patterns and create predictions,
The procedure in degradation of the pollutants that gets attached on machine learning can be used to examine huge datasets, including
to photocatalyst undergoes redox reaction that involves charge transfer experiment findings. It can be used, for instance, to spot patterns in the
to produce the hydroxyl radicals and superoxide anions [15]. The hy­ behavior of molecules, forecast the characteristics of novel chemicals, or
droxyl and O−2 radicals are generated on the surface of the photocatalyst group substances according to their chemical makeup. Machine learning
by the valence and conduction bands such that the holes in the valence can also be used to create novel catalysts, anticipate how new chemicals
band capture the water from the solution and form the hydroxyl radicals will react, and improve the synthesis of molecules. In chemical systems,
and on the other hand the excited electrons in the conduction bands machine learning can be used to identify and diagnose anomalies, such
reacts with the oxygen to form the superoxide radicals both of these as the presence of pollutants or dangerous compounds. There are a va­
radicals are oxidizing agents which plays the crucial role in the degra­ riety of machine learning tools reported in literature each having unique
dation of the organic dyes. This process of the formation of the radicals characteristics. This study utilizes Artificial Neural Networks for the
allows charge separation to be protracted [16–18]. prediction of bandgap of transition metal doped ZnO.
Over the course of time, the use of ML and AI methods has exten­
sively increased for studying chemical dynamics and properties of ma­ 2.1. Neural networks
terials. Training of large datasets using ML tools allow for an efficient
and dynamic analysis for the quality-structure property relationships A branch of machine learning and artificial intelligence called neural
(QSPR) of chemical compounds. These techniques allow for establishing networks makes an effort to imitate how the human brain works in order
an empirical relationship between variables enabling the system for a to solve challenging problems. They are made up of interconnected
very accurate and precise calculation of chemical properties. The nodes or neurons that use mathematical algorithms to analyze and
advantage of these methods is the estimation of properties using data transmit information. This enables them to recognize patterns in data,
available in literature and optimization of reaction parameters for learn from it, and build hypotheses or come to conclusions based on the
computational studies. These techniques allow for development of re­ patterns they have discovered. The first mathematical model of a
lationships between parameters of simplistic nature with those that simplified neuron was presented by Warren McCulloch and Walter Pitts
can’t be obtained without experimental analysis. This serves as an in 1943, and this is where the concept of ANNs first emerged. Based on
excellent computational tool for future synthesis studies while also the work of Frank Rosenblatt, who presented the idea of a single-layer
having great significance in property estimation studies. neural network capable of solving linearly separable problems, the
This study focuses on developing a hybrid method for the estimation concept of perceptrons emerged in the late 1950s and early 1960s.
of Bandgap of transition metal doped Zinc oxide using techniques based However, as a result of the perceptron’s shortcomings in tackling
on neural networks. When studying the optical properties of these complicated issues that were not linearly separable, interest in ANNs
photocatalytic materials, it is essential to determine their bandgap as its began to fade in the 1970s. The 1980s and 1990s saw a resurgence in

2
M.H. Saeed et al. Chemometrics and Intelligent Laboratory Systems 242 (2023) 104983

interest in ANNs thanks to the development of backpropagation, a FANNs are able to learn from data and can be applied to data classifi­
potent multi-layer neural network training technique. Multiple re­ cation or prediction.
searchers separately created backpropagation, which allowed ANNs to In multiple industries, including engineering, finance, and medicine,
effectively learn from data with multiple layers and circumvent the ANFIS has found use. Here are a few literary examples of ANFIS
drawbacks of single-layer perceptrons. The revival of ANNs and the applications:
broader field of machine learning occurred throughout the course of the
next decades as a result of considerable improvements in computer • Predicting energy consumption: ANFIS has been used to predict the
power, the accessibility of large datasets, and the development of energy consumption of buildings, based on factors such as outdoor
increasingly sophisticated algorithms. The widespread adoption of temperature, time of day, and building occupancy. This can help to
graphical processing units (GPUs) further accelerated the training of optimize energy usage and reduce costs [19].
deep neural networks, enabling the development of Deep Learning, a • Financial forecasting: ANFIS has been used to predict stock prices
subfield of machine learning that uses multi-layer neural networks to and exchange rates, based on historical data and market trends. This
solve complex problems. can help investors make more informed decisions and reduce risks
These networks can be applied to a wide range of tasks, including [20].
speech recognition, language translation, and image recognition. Ro­ • Medical diagnosis: ANFIS has been used to diagnose diseases such as
botics, computer vision, and natural language processing have all diabetes and cancer, based on patient data such as age, gender, and
advanced significantly as a result of neural networks’ capacity to learn medical history. This can help to improve the accuracy and speed of
from and adapt to vast datasets. As a result, neural networks are now a diagnosis [21].
crucial component of contemporary technology and are still being • Fault detection in industrial systems: ANFIS has been used to detect
actively researched and developed. The network can be used to make faults in industrial systems, such as pumps and turbines, based on
predictions based on brand-new, undiscovered data after being trained. sensor data and system performance. This can help to improve
Neural networks have been used in chemistry for a variety of purposes, maintenance and reduce downtime [22].
including drug discovery and chemical process prediction. Following are • Traffic flow prediction: ANFIS has been used to predict traffic flow
some possible applications for neural networks in chemistry: and congestion, based on data such as weather conditions, road
conditions, and historical traffic patterns. This can help to optimize
• Drug discovery: Neural networks can be used to predict the proper­ traffic management and reduce congestion [23].
ties of new molecules, including their toxicity, bioavailability, and
potential as drug candidates. This can help researchers identify These are only a few of the numerous uses for ANFIS in different
promising new drug candidates more quickly and efficiently. industries. A powerful and adaptable method for data analysis and
• Molecular modeling: Neural networks can be used to simulate the decision-making, ANFIS has shown numerous times, and it is likely to be
behavior of molecules and predict their properties, such as solubility a very important study area in the years to come.
and stability. This can be useful in designing new materials, opti­
mizing chemical processes, and understanding the behavior of 3. Data collection
complex chemical systems.
• Spectroscopy: Neural networks can be used to analyze spectroscopic Transition metal doping of zinc oxide can affect its electronic and
data, such as infrared and nuclear magnetic resonance spectra, to optical properties, including its band gap energy. The exact effect of
identify unknown compounds and determine their chemical doping on the band gap depends on several factors, such as the type of
structure. transition metal, its concentration, and the crystal structure of the ma­
• Reaction prediction: Neural networks can be trained to predict the terial. Considering the transition metal dopants on ZnO, each of them
outcomes of chemical reactions based on the input of reactants and has a particular effect on the bandgap resulting in a unique bandgap.
reaction conditions. This can be useful in designing new synthetic Most of the factors that are responsible for the changes in the bandgap
routes and optimizing chemical processes. are the same among different periods of the periodic table. The values
• Materials design: Neural networks can be used to predict the prop­ for the atomic radii and the ionic radii were taken from Clementi et al.,
erties of new materials, such as their conductivity, strength, and and Slater et al., [24,25]. Here, the data is divided into 3 classes based on
thermal stability. This can be useful in designing new materials for the periods of the periodic table from which the dopants belong to.
specific applications, such as electronics, energy storage, and
catalysis. 3.1. Period 3 transition metal dopants

Overall, through facilitating quicker and more effective drug dis­ To start off with period 3 transition metal dopants, scandium doping
covery, materials design, and chemical process optimization, neural on ZnO thin film led to an increase in the bandgap from 3.29 to 3.31 eV
networks have the potential to revolutionize chemistry. with an increase in dopant concentration from 0 to 1.0 wt%. The
observed increase as reported was caused by the increase in the grain
2.2. Fuzzy artificial neural networks (FANNs) size leading to greater scattering centers leading to a decrease in optical
transmission. A loss in optical transmission is usually observed with the
An ANN type called fuzzy artificial neural networks (FANNs) uses loss of light because of oxygen vacancies and the scattering phenomenon
fuzzy reasoning to handle input. In order to combine the precision and observed at the grain boundaries [26]. Titanium doped on ZnO lead also
accuracy of neural networks with the durability and interpretability of lead to an increase in the average visible transmittance allowing for
fuzzy logic, FANNs are used to describe complicated nonlinear systems. 88.7% maximal value at 3 wt% with 86.5% at 5 wt%. The observed
Pattern recognition, classification, and control applications are where bandgap of the 3 wt% was also greater than that of pure ZnO because of
FANNs are most frequently utilized. Additionally, they can be applied to Burstein–Mossshift [27]. Similar effects on the bandgap have been
grouping, optimization, and forecasting. A group of interconnected observed in literature that explain the effect in Gallium and Aluminum
nodes, or neurons, that are arranged into layers make up FANNs. The doped ZnO films [28,29]. Vanadium doping also follows the same
inputs and outputs of each neuron are combined to create a fuzzy rule. pattern of increasing the bandgap but with a slight deviation at lower
The fuzzy rule is then used to calculate the neuron’s output. The concentrations as illustrated in Table 1. This maybe because of sp-d
disparity between the actual output and the anticipated output is exchange interactions taking place among the band on c-axis of ZnO
minimized by training FANNs by changing the weights of the neurons. and Bernstein-Moss Effect [30,31].

3
M.H. Saeed et al. Chemometrics and Intelligent Laboratory Systems 242 (2023) 104983

Table 1
Literature review of period 3 transition metal doped zinc oxide.
Composite Element Bandgap Concentration of Grain Size Performance Measure Dye Degraded Synthesis Method Ref
Name Doped (eV) dopant (/wt%)

Sc @ZnO Sc 3.31 1 25 – – Sol-gel process [26]

3.3 0.5 22
3.29 0 21
Ti @ZnO Ti 3.52 5 62 – – RF Magnetron [27]
3.49 3 80 Sputtering
3.52 2 72
V @ ZnO V 3.055 10 61.2 – – Ceramic Double [39]
3.063 8 59.3 Sintering Method
3.072 6 59.6
3.083 4 59.3
3.126 2 61.2
3.258 0
Cr @ZnO Cr 3.34 3 30–40 Enhanced photo degradation Methylene Blue Sol–gel technique. [32]
3.38 2 30–41 efficiency (75.5%) compared to
3.43 1 30–42 pure zinc oxide nanoparticles
3.49 0 (51.2%)
Mn @ZnO Mn 3.51 4 28.6 87/93.5% (MO), 85/88% (MB), Methylene Blue, Congo Sol-gel Method [40]
3.36 0 30.61 and 86/93% (CR) Red and Methyl
(Pure ZnO/Mn@ZnO) Orange
Fe @ZnO Fe 3.17 Nw 50-100 – Methyl orange and 1,2- Hydrothermal [41]
nanowires diameter dichlorobenzene Method
(hexagonal)
3.18 Pure ZnO 30–50 nm
3.26 Np’s
Fe@ZnO Fe 3.24 7.5 20–50 71% color removal, 94% dye Acid Red Combustion [42]
3.19 5 20–50 degradation, and 41% total Synthesis Method
3.14 2.5 20–50 organic carbon removal for acid
3.23 1 20–50 red 1 dye solution
Co@ZnO Co 3.65 10 12.2 8 mol % Co-doped ZnO, 98% of Methylene Blue Chemical [43]
3.7 8 13 the MB dye in 180 min under Precipitation
3.76 6 13.6 visible light irradiation. Method
3.78 4 14.7
3.79 2.5 15.3
3.8 0 15.8
Ni@ZnO Ni 3.14 20 43 Best performance shown by 10% Rhodamine B Simple water bath [44]
3.18 15 31 Ni@ZnO (92.2% under the UV method
3.19 12.5 45 light irradiation for 210min)
3.21 10 34
3.24 7.5 50
3.25 5 21
3.26 2.5 46
3.28 0
Cu @ZnO Cu 2.94 7 37.8 – – Chemical Co- [45]
2.95 5 18.95 precipitation
2.99 2 19.61 Method
3.01 1 21.62

Chromium had similar red shift effects over the bandgap as a result of Also, the redshift can be explained by the strong interactions between p-
an emerging impurity in the conduction band of ZnO and with the in­ d mixing of oxygen and the Cu2+ ions [37]. Bylsma et al. used the second
crease in dopant concentrations, it kept on decreasing [32]. Manganese order perturbation theory to explain the shifting of the bandgap using
doping allowed for increase in bandgap of ZnO because of the replace­ s-d and p-d exchange interactions [38].
ment of Zn2+ with Mn2+ ions in the crystal lattice [32]. Burstien-Moss
Effect may be responsible for the rise in the bandgap of ZnO in the 3.2. Period 4 transition metal dopants
case of Iron at higher dopant concentrations of 5 and 7.5%. While lower
concentrations of iron got blocked due to closing of distance between the Yttrium doping also seems to be a viable option when considering the
fermi bands and the conduction band because of an increase in the narrowing of bandgap. Yttrium doping led to a decrease in the bandgap
carrier concentration causing the value of Eg to decrease [33]. The of around 80 meV. It can be pointed out that the increase in doping
decrease in bandgap energy in case of Cobalt is bestowed upon the ex­ concentration and decrease in size lead to a decrease in bandgap. In this
change of “sp-d” and localized “d” electron of the newly introduced scenario, the Y3+ act as electron donors in the ZnO lattice, and the donor
Cobalt dopant in the lattice [34,35]. Nickel doping also led to a decrease energy of the ions is on par with the conduction band. The greater ionic
in the bandgap as it replaced the Zinc atoms in the ZnO lattice narrowing radius of the Y-ion induces the donor wave functions of electrons from Y
the bandgap. Also, the concentration of the dopant was more significant in the neighboring lattice to overlap. Increasing the concentration of the
than the Mott transition concentration allowing for a lower bandgap impurity leads to more electronic overlap splitting the donor energy
according to the Theory of Semiconductor-metal transition [36]. level broadening the impurity band [46].
In case of copper, minute changes in bandgap are observed in the Zirconium doping led to the generation of two electrons when it
synthesized nanoparticles when Cu2+ ions are replaced by Zn2+ in the replaced Zinc ions in the lattice of ZnO which led to an increase in the
crystal lattice. This change in bandgap (red shift) could be explained by electrical conductivity. Also, it was noted that the structures are oxygen
p-d spin exchange interactions between the localized d electrons and the deficient, so it was clear that the carrier sources are oxygen vacancies
band electrons of the substituting transition metal atom in the lattice. and Zirconium doping. The doping lead to a significant decrease in the

4
M.H. Saeed et al. Chemometrics and Intelligent Laboratory Systems 242 (2023) 104983

number of oxygen vacancies which were similar to Aluminum doped holes increasing its photocatalytic performance. Even though the
ZnO films already reported in literature [47]. The photocatalytic study bandgap is lower in case of 1% Nb doping, the 2% doped nanoparticles
of Niobium doped Zinc Oxide nanoparticles revealed an increase in performance was better due to symmetric lattice spacing and structural
bandgap due to Moss-Burstein effect. The 2% Nb doped ZnO allowed for uniformity [48].
a significant increase in the charge separation between the electrons and The transparent hexagonal polycrystalline molybdenum doped thin

Table 2
Literature review of period 4 transition metal-doped zinc oxide.
Composite Element Bandgap (eV) Concentration of Grain Size Performance Measure used in the Dye
Name Doped dopant (/wt%) article

Degraded Synthesis Ref

Method
Y @ ZnO Y 3.25 15 23 – – Hydrothermal- [46]
Precipitation
Method.
3.2 9 40
3.17 6 25
3.21 3 28
3.24 1 32
3.25 0 42
Zr@ZnO Zr 3.23 10 16.6 – – Radio Frequency [54]
Magnetron
Sputtering
3.35 5 20
3.32 0 11.1
Nb @ZnO Nb 3.19 0.025 33.84 The UV-light induced photo catalytic Methylene Sol-Gel Method [48]
activity of 2% Nb-doped ZnO is found Blue
to be twice that of undoped ZnO.
3.18 0.02 34.12
3.16 0.01 34.43
3.15 0.005 37.83
3.09 0 39.62
Mo@ZnO Mo 3.8 3 35 – – DC Reactive [49]
Magnetron
Sputtering
3.72 2.3 30
3.66 1.7 27
3.57 1.2 25
3.5 0.6 26
3.37 0
Tc@ZnO No work was
found in the
literature for Tc
doped Zinc
Oxide.
Ru @ZnO Ru Band gap was Broad Gas sensing Sol–gel [55]
unchanged absorption Method
peak centered
at 800 nm.
3.27 0
Rh @ZnO Rh Work is done by – – – – –
can’t find the
bandgap
Pd @ZnO Pd 3.25 Microwave 30.1 The photodegradation percentage of Congo Red Microwave- [50]
Hydrothermal dye is about 66%, 91%, 95.1% and hydrothermal
Method 98.2% for ZnO, Pd/ZnO-1, Pd/ZnO-2 methods
and Pd/ZnO-3, respectively after UV
irradiation for 120 min
3.25 Microwave 31.33
Method
3.25 Photo-Reduction 33.46
Method
3.25 Borohydride 34.43
Reduction
Ag @ZnO Ag 3.1 eV * ZnO:Ag nanoparticles exhibited 3% Tryptan Low temperature [51]
higher photocatalytic efficiency than Blue combustion
pure ZnO nanoparticles. synthesis
Ag @ZnO Ag 3.207 8 18.4 [35]
3.229 1 19.57
3.237 0.3 20.29
3.241 0 22.06
Cd @ZnO Cd 3.06 20 400–600 5% Exhibited the best performance Rhodamine Hydrothermal [53]
B Method
3.12 10
3.16 5
3.24 0

5
M.H. Saeed et al. Chemometrics and Intelligent Laboratory Systems 242 (2023) 104983

films revealed that the doping allowed for the existence of Molybdenum charge carrier specie results in a a blue shift [61]. Tungsten-doped Zinc
in multiple oxidation states, namely Mo5+ and Mo6+. This allowed for oxide thin films was synthesized using a pulsed laser deposition method
addition of extra electrons in the lattice without using higher doping for examination of their optical and structural properties. A small
concentrations. It was reported that the absorption edges shifted on decrease in the optical bandgap was observed with the increase in the
lower wavelength with the increase in doping concentrations [49]. laser fluence [62]. This minor reduction is due to the increased thickness
Extensive literature review didn’t reveal any works related to techne­ of the film and higher crystallinity due to the influence of higher laser
tium doped zinc oxide as it was thought to have no effect on the bandgap fluence [63]. After extensively searching for osmium and
or the photocatalytic activity of zinc oxide. Ruthenium doping was rhenium-doped zinc oxide nanoparticles or thin films, no records were
effective in terms of improving the photocatalytic activity of zinc oxide, found. Iridium-doped zinc oxide nanoparticles with a variable doping
but optical studies revealed no effect of the properties of the bandgap of concentration of 5, 10, and 15% revealed a bandgap of 2.72, 2.72, and
the composite. With this in mind, a broad peak with a center at 800 nm 2.71 eV, respectively. No significant changes in the bandgap were
was observed which is likely the reason for the behavior of the material observed after 5 wt% doping suppressing electron-hole pairs’ recombi­
in visible light [48]. The experimental results of Palladium doping on nation ability, making it a good photocatalyst [64]. Platinum-doped zinc
Zinc Oxide using different doping methods (Microwave Hydrothermal oxide nanoparticles showed a red shift with the absorption edge shifting
Methods, Microwave Method, Photo-Reduction Method, and Borohy­ from 387 to 468 nm. The bandgap decreased from 3.2 to 2.76 eV with
dride Method) from a study by Güy et al., indicated no change in optical the increase in dopant concentration from 0.1 to 0.4 wt% [64]. No
bandgap properties maintaining the initial bandgap of 3.25 eV [50]. significant decrease was found after 0.3 wt% and the findings corre­
Ravishankar et al., prepared Silver doped zinc oxide nanoparticles using spond to those already reported in the literature [65]. This decrease in
a low-temperature combustion synthesis method. The exploration of the bandgap is due to the incorporation of platinum ions on the surface of
optical properties of the silver doped zinc oxide nanoparticles indicated zinc oxide, lowering the bandgap. A complete literature review of
a red shift with a bandgap of 3.100 eV from the 3.370 eV of the pure ZnO transition metal dopants is provided in Table 3, given below.
nanoparticles. The silver nanoparticles have greater absorption in
comparison to zinc oxide causing the red shift [51]. 4. Methodology
A study of similar nature of preparation of Silver doped zinc oxide
nanoparticles by Yildirim et al., revealed similar results of decrease in This section of the article provides a brief explanation of the methods
bandgap from 3.241 eV to 3.207 eV at 8% Ag concentration. This is due that are employed for the estimation of transition metal doped ZnO.
to several factors like the absorption edges, the size, shape, dopant
concentration, and the defects. Studies from literature reveal the gen­ 4.1. Multi-layer percepteron
eration of oxygen vacancies when an electron is excited from the con­
duction to the valence band causing a decrease in the bandgap [52]. For The motivation for the design of an MLP system comes from the
Cadmium doping, a slight decrease in the bandgap was observed as the complex nervous system of living organisms. The system is composed of
Cd2+ ions replaced the Zn2+ ions in the lattice of the nanoflowers [53]. A numerous parallel processing units interconnected using linkages
complete literature review of transition metal dopants is provided in assigned with different weights [67]. Each of the inputs is assigned an
Table 2 given below. appropriate weight and the sum of the weighed inputs serve as the input
for the transfer function f. The system can use as many transfer functions
3.3. Period 5 transition metal dopants as required for continuous input weight correction for the training
process until the trained data can correctly predict the output values.
Lanthanum doping of Zinc oxide does yield interesting results in The process involves the use of known outputs to determine the errors
terms of its effect on the bandgap of the material. Lanthanum doping used for the estimation process. The errors are backpropagated and
lowered the material’s bandgap from 3.1 to 2.78 eV by increasing the incorporated in the weight adjustment until minimum error is acquired.
doping concentration from 0 to 10 mol%. These results are similar to The Levenberg-Marquardt method is used for calculating the weights for
those in literature for lanthanum-doped titanium dioxide [56]. The low the MLP [68]. Mathematically, the relation between the input and
decomposition temperature of the nitrate (126◦ ) allowed for the for­ output is expressed by the equation:
mation of La2O3 in the zinc oxide lattice allowing for La3+ ions to enter The output is expressed as a function (f(n)) of n where:
the zinc oxide lattice during the annealing process easily. The La3+ can
now be situated in the place of the Zn2+ ions in the lattice structure or
∑
X
n= wj .rj + P, …….. r1 , r2 , r3 … rX , (1)
can be found in the interstitial voids [57]. Hafnium-doped zinc oxide j=1
nanoparticles through the sol-gel method are reported in the literature.
The bandgap was reported to decrease from 3.26 to 3.17 eV with the In this equation, r1, r2, r3, …rN are inputs, P is the bias function, while
increase in Hafnium concentration from 0 to 6%. A possible reason for the weights for each of the neurons are represented by w. For the acti­
this change in bandgap is bandgap edge bending [58]. This change can vation function, linear and sigmoid functions were utilized. The results
be due to the substitution of Zn2+ with Hf4+ allowing for sp-d exchange of the models made using the Multi-Layer Perceptron approach are
interactions between the band electrons of the localized d electrons of expressed in Table 4.
the Hf4+ ions [59].
Tantalum doping of zinc oxide revealed variating results as the 4.2. Adaptive neuro-fuzzy interference system (ANFIS)
bandgap increased from 3.27 to 3.38 eV with an increase in Ta con­
centration from 0 to 5%. The bandgap then decreased to 3.34 when the With the use of an Adaptive Neuro-Fuzzy Inference System (ANFIS),
dopant concentration increased to 10 wt%. This change is due to the fuzzy-based rules may be incorporated into neural networks, generating
Moss-Burstein effect which states that the charge-carrying concentration a very potent tool for training inaccurate and incomplete data [69].
increased with the doping of tantalum, allowing for the energy states in Jyh-Shing Roger Jang developed ANFIS in the early 1990s with the goal
the bottom of the conduction band to fill up completely, increasing the of developing a flexible system that could seamlessly handle both lin­
fermi level higher to the conduction band. With this, the semiconductor guistic and numerical input. ANFIS’s architecture is made up of a
nature changed from non-degenerate to degenerate [60]. In order for the number of layers, each of which has a distinct purpose. Using mem­
electrons valence electrons to become a free charge carrier, they must be bership functions, the first layer, also referred to as the "fuzzification
excited above the fermi level resulting in the increase of the bandgap. A layer," converts the input data from crisp to fuzzy sets. The degrees of
similar effect in doped thin films is observed where the increase in the membership of the fuzzy sets are combined in the second layer, the "rule

6
M.H. Saeed et al. Chemometrics and Intelligent Laboratory Systems 242 (2023) 104983

Table 3
Literature review of period 5 transition metal doped zinc oxide.
Composite Element Bandgap Concentration of dopant Grain Dye Degradation Efficiency Dye Synthesis Method Ref
Name Doped (eV) (/wt%) Size Degraded
(nm)

La @ZnO La 2.78 10 10.3 85.86% after 150 min Methyl Gel-Combustion [57]
2.82 5 12.1 Orange Method
2.91 1 18.9
3.1 0 34.3
Hf@ZnO Hf 3.17 6 16 2 mol% Hf-doped ZnO showed best activity Methylene Sol-Gel Method [59]
3.2 4 15 by reducing MB concentration by 85% in Blue
3.215 3 15 120min
3.23 2 14
3.245 1 15
3.26 0 17
Ta @ZnO Ta 3.342 10 13.7 The average transmittance in the visible – Magnetron [60]
3.36 7 13 areas is over 85%. The maximum optical Sputtering
3.38 5 11.8 band gap is 3.38 eV Apparatus.
3.29 2 11
3.25 0 10.5
W @ZnO W 3.31 1% wt, variation of laser 581 – – Pulsed Laser [62]
3.34 fluence is 2.7,1.7,1 501 Deposition
3.36 respectively 434
3.37 0
Re@ZnO – – – – – – –
Os@ZnO – – – – – – –
Ir@ZnO Ir 2.71 15 33 Maximum MG removal was obtained at Malachite Sono-chemical [64]
2.72 10 30 0.1 ppm MG concentration, 0.332 mmol Green method
2.72 5 39 K2S2O8, and 5% Ir doped ZnO
3.2 0 95
Pt@ZnO Pt 2.76 0.4 27 0.3 wt % Pt showed best photocatalytic Malachite Sol-gel Method [66]
2.65 0.3 27 activity Green
2.85 0.2 27
2.97 0.1 30
3.2 0 26

Table 4
Performance Statistics of MLP based models.
Model No. Inputs Model Tag No. of Epochs Hidden Layers Training Phase Data Testing Phase Data

RMSE MAE R RMSE MAE R

1. AN MLP(I1) 37 3 0.1395 0.0793 0.463 0.19465 0.1296 0.2182

2. GS MLP(I2) 22 3 0.2238 0.1680 − 0.1384 0.30239 0.24784 0.0311
3. AR, IR MLP(I2,2) 22 6 0.1179 0.0550 0.3074 0.26685 0.20002 0.3505
4. GS, C MLP(I2,4) 36 9 0.2027 0.1525 0.3355 0.3057 0.2463 0.3141
5. AN GS, C, AR MLP(I3,2) 38 25 0.2073 0.1340 0.7546 0.3055 0.2444 0.6188
6. AN, GS, C, IR MLP(I4,4) 44 2 0.1517 0.1023 0.4191 0.3559 0.2509 0.04120
7. AR, IR, AN, C, GS MLP(I5,5) 46 3 0.2033 0.1232 0.6593 0.2765 0.2176 0.2692

layer," to determine the firing strength of each rule. The weighted performed better in terms of accuracy, adaptability, and efficiency. To
average of the rule firings is used in the third layer, also known as the further improve ANFIS’s capabilities, it has been coupled with other
"consequent layer," to calculate the output. The "defuzzification layer," methodologies like genetic algorithms, swarm intelligence, and deep
the fourth layer, transforms the fuzzy output back into a crisp value. A learning [71–75].
mathematical framework known as fuzzy logic uses linguistic variables Here, ANFIS was used for optimal model training using a variety of
and the rules that go along with them to deal with uncertainty and combinations of the inputs, spread functions, and membership functions
imprecision. It is especially helpful in circumstances when conventional (i.e., triangular, gaussian etc.). The results of the model are illustrated in
binary logic may not be suitable since it enables a machine to reason and Table 5.
make judgments using "if-then" rules that are similar to those used by
humans. ANNs, on the other hand, are potent machine learning models 5. Results and discussion
that draw inspiration from the neural organization of the human brain.
They are adept at identifying patterns and roughly simulating complex To model the bandgap of the doped ZnO nanomaterials, the litera­
functions from input-output data pairs, which makes them suitable for a ture was thoroughly studied, and data samples were collected each
variety of jobs like pattern recognition, classification, and regression. having values of the variables discussed in Section 3. The data was
ANFIS successfully combines the advantages of ANNs and fuzzy logic divided into two parts training and testing (70% and 30%, respectively).
to produce a highly flexible and intelligent system. The core concept of The models were trained to get the optimal number of epochs, hidden
ANFIS is to transform fuzzy rule bases into a neural network structure, layers, fit functions, and learning rates for the development of the best
and then to adaptively optimize the network’s parameters to produce architecture of the neural networks for the model training process. Each
precise and adaptable outputs [70]. Numerous domains, including sys­ of the models was trained so minimal error was achieved during the
tem identification, prediction, control, and pattern recognition, have process of generating the predicted outcome. A variety of codes were
effectively used ANFIS. When compared to other models, it has written in MATLAB (Neural Network Toolbox), and the accuracy of the

7
M.H. Saeed et al. Chemometrics and Intelligent Laboratory Systems 242 (2023) 104983

Table 5
Results of Performance Statistics of ANFIS based models.
Model No. Inputs Model Tag Optimization Model Membership Function Training Phase Data Testing Phase Data

1. AN ANF(I1) Hybrid Optimization model Triangular membership functions RMSE RMSE

0.218069 0.268069
2. GS ANF(I2) 0.233242 0.283242
3. AR, IR ANF(I2,2) 0.201663 0.281663
4. GS, C ANF(I2,4) 0.225035 0.285035
5. AN, GS, C ANF(I3,2) 0.185686 0.335686
6. AN, GS, C, AR ANF(I4,4) 0.0937076 0.1237076
7. AR, IR, AN, C, GS ANF(I5,5) 0.0242682 0.0342682

models was determined using statistical measures (Root Mean Square Compared to the results generated from the MLP models, the per­
Value (RMSE) and the Mean Absolute Error (MAE). formance of those models that contain lesser number of variables (i.e.,
The results of the performance statistics of the MLP models in Table 4 MLP(I1), MLP(I2), MLP(I2,2), and MLP(I2,4)) performed significantly
indicate low RMSE values during both the testing and training phases better indicated by the lower RMSE values. But with the increase in the
indicating superior performance for predicting the bandgap for each of number of inputs for the bandgap prediction, ANFIS models performed
the dopants on ZnO. A low RMSE means that the differences between the better due to increased number of membership functions and the
predicted values and actual values of the target variable are small, incorporation of fuzzy logic in the prediction process. ANFIS can deal
implying that the model is performing well in capturing the underlying with imprecise and imperfect data, making it more suitable when the
patterns and relationships in the data. Out of all the models designed number of variables is greater.
using different variables, MLP(I1), MLP(I2,4), and MLP(I5,5) had the Current computational studies involve the use of Density Functional
lowest RMSE values alongside other statistics allowing for the best Theory and other associated methods for estimation of bandgap which
prediction of the bandgap of the semiconductor. are both time intensive and computationally burdening tasks. The above
The model fitting for each of the variables was explored using fitting models can serve as a precise synthesis and computational tool that can
curves generated using the Matlab nntool. The statistical findings predict the properties of a material beforehand based on the desired
coincide with the graphical findings as the relationship met the training property of the material that is to be synthesized. This will help a great
targets developed by the MLP models. The best relationship between the deal in determination of optical properties without any sort of experi­
input and output variables are illustrated in model MLP(I5,5) and MLP mental lab work.
(I3,2). The regression plots for MLP(I5,5) and MLP (I3,2) are given in The prediction of changes in the band gap of doped semiconductors
Figs. 1 and 2. using artificial intelligence (AI) has significant implications for materials
The fit function can be interpreted as a mathematical representation science and electronics engineering. Traditionally, the prediction of the
of the relationship between the input features and the target variable band gap of doped semiconductors has been a time-consuming and
(see Fig. 3). For example, in linear regression, the fit function is a linear resource-intensive process that relies on experimental measurements
equation that describes the relationship between the input features and and theoretical calculations. However, AI algorithms can be trained on
the target variable. The fit function curve for MLP(I5,5) is illustrated in large datasets of experimental and computational data to learn the
Fig. 1. complex relationships between the chemical composition, crystal
The ANFIS models were trained using the Fuzzy Logic Toolbox in structure, and electronic properties of doped semiconductors. Once
Matlab, and the performance measures for each of the models are re­ trained, these AI models can predict the band gap of novel or modified
ported in Table 5. The models were trained using constant-type trian­ doped semiconductors accurately and quickly, accelerating the search
gular membership functions, while a hybrid optimization model was for new electrical and photonic materials.
selected for the model training purposes. The division of the dataset was AI-based band gap prediction for doped semiconductors can also be
done in a 70-30 ratio, training and testing, respectively. The results used to improve the design of already-existing electrical and photonic
suggest a great model fitting with a decrease in RMSE values when the devices. AI models can boost the effectiveness and performance of solar
number of variables for bandgap prediction increases, indicating better cells, LEDs, and other electrical devices, for instance, by forecasting the
performance. ideal dopant concentration or chemical composition for a particular

Fig. 1. Regression plot of MLP(I3,2).

8
M.H. Saeed et al. Chemometrics and Intelligent Laboratory Systems 242 (2023) 104983

Fig. 2. Regression plot of MLP(I5,5).

Fig. 3. Fit function for MLP(I5,5).

application. optical, and functional properties of doped substances. The current

In conclusion, the use of AI to forecast changes in the band gap of models are to be expanded to determine their performance using the
doped semiconductors has important ramifications for materials science current and additional parameters to determine the photocatalytic, anti-
and electronics engineering, enabling the development of novel mate­ bacterial, anti-fungal, and electrochemical properties for doped zinc
rials and the improvement of current devices. oxide nanomaterials. The prediction of changes in the band gap of doped
semiconductors using AI has significant implications for materials sci­
6. Conclusion ence and electronics engineering, enabling the discovery of new mate­
rials and the optimization of existing devices.
An ANFIS and Multi-Layer Perceptron-based model was constructed
for the analysis and prediction of the optical bandgap of transition metal Author statement
doped zinc oxide nanomaterials using basic parameters. The study
aimed to develop a system that can serve as an efficient guide for future Muhammad Haris Saeed write and edit the manuscript, designed the
synthesis for determining properties using simple parameters just by mathematical model and algorithms in the manuscript for the consid­
using data obtained from the literature. The results were highly prom­ ered problem, Naveen Kosar write and edit the manuscript, designed the
ising as the trained models were highly accurate in predicting the mathematical model and algorithms in the manuscript for the consid­
bandgap using all five factors considered in the study. The technique has ered problem, Sadaf-ul-Hassan write and edit the manuscript, designed
promising potential for developing libraries serving as guides using data the mathematical model and algorithms in the manuscript for the
from the literature for prediction analysis of different physiochemical, considered problem, Sohail Nadeem write and edit the manuscript,

9
M.H. Saeed et al. Chemometrics and Intelligent Laboratory Systems 242 (2023) 104983

Fig. 4. RMSE of the trained Models.

designed the mathematical model and algorithms in the manuscript for [8] A. Sobczyński, A. Dobosz, Water purification by photocatalysis on semiconductors,
Pol. J. Environ. Stud. 10 (2001) 195–205.
the considered problem, Mazin Abed Mohammed: Formal analysis,
[9] D. Bahnemann, Photocatalytic water treatment: solar energy applications, Sol.
Methodology, and project administration, Mohd Khanapi Abd Ghani Energy 77 (2004) 445–459.
Formal analysis, Methodology, Karrar Hameed Abdulkareem Formal [10] R. Molinari, M. Borgese, E. Drioli, L. Palmisano, M. Schiavello, Hybrid processes
analysis, Methodology (see Fig. 4) coupling photocatalysis and membranes for degradation of organic pollutants in
water, Catal. Today 75 (2002) 77–85.
[11] M.D. Hernández-Alonso, F. Fresno, S. Suárez, J.M. Coronado, Development of
Funding alternative photocatalysts to TiO 2: challenges and opportunities, Energy Environ.
Sci. 2 (2009) 1231–1257.
[12] D. Ravelli, D. Dondi, M. Fagnoni, A. Albini, Photocatalysis, A multi-faceted concept
This research received no external funding”. for green chemistry, Chem. Soc. Rev. 38 (2009) 1999–2011.
[13] K. Soutsas, V. Karayannis, I. Poulios, A. Riga, K. Ntampegliotis, X. Spiliotis,
Declaration of competing interest G. Papapolymerou, Decolorization and degradation of reactive azo dyes via
heterogeneous photocatalytic processes, Desalination 250 (2010) 345–350.
[14] A. Primo, A. Corma, H. García, Titania supported gold nanoparticles as
The manuscript is an original work on its merit, that it has not been photocatalyst, Phys. Chem. Chem. Phys. 13 (2011) 886–910.
previously published in whole or in part, and that it is not being [15] W.Y. Teoh, J.A. Scott, R. Amal, Progress in heterogeneous photocatalysis: from
classical radical chemistry to engineering nanomaterials and solar reactors, J. Phys.
considered for publication elsewhere. Also, all authors have read the Chem. Lett. 3 (2012) 629–639.
final manuscript, have approved the submission to the journal, and have [16] T.K. Tseng, Y.S. Lin, Y.J. Chen, H. Chu, A review of photocatalysts prepared by sol-
accepted full responsibilities for the manuscript’s delivery and contents. gel method for VOCs removal, Int. J. Mol. Sci. 11 (2010) 2336–2361.
[17] M.A.A. Henderson, Surface science perspective on TiO2 photocatalysis, Surf. Sci.
In addition, the authors declare that there is no conflict of interests.
Rep. 66 (2011) 185–297.
[18] K. Pirkanniemi, M. Sillanpää, Heterogeneous water phase catalysis as an
Data availability environmental application: a review, Chemosphere 48 (2002) 1047–1060.
[19] A. Khan, M. Rizwan, ANN and ANFIS based approach for very short term load
forecasting: a step towards smart energy management system, in: Proceedings of
Data will be made available on request. the 2021 8th International Conference on Signal Processing and Integrated
Networks (SPIN), IEEE, 2021, pp. 464–468.
Appendix A. Supplementary data [20] W. Lu, J. Li, J. Wang, L. Qin, A CNN-BiLSTM-AM method for stock price prediction,
Neural Comput. Appl. 33 (2021) 4741–4753.
[21] D. Ezzat, A.E. Hassanien, H.A. Ella, An optimized deep learning architecture for the
Supplementary data to this article can be found online at [Link] diagnosis of COVID-19 disease based on gravitational search optimization, Appl.
org/10.1016/[Link].2023.104983. Soft Comput. 98 (2021), 106742.
[22] A. Bakhtiar, I.W. Nurwidanto, S. Hartini, P.A. Wicaksono, Process improvement
design at PT URW using failure mode and effect analysis, Proc. IOP Conf. Ser.:
References Earth [Link]. 896 (2021), 12077. IOP Publishing.
[23] F. Zhou, L. Li, K. Zhang, G. Trajcevski, Urban flow prediction with spatial–temporal
[1] M.A. Qamar, S. Shahid, M. Javed, Synthesis of dynamic G-C3N4/Fe@ZnO neural ODEs, Transport. Res. C Emerg. Technol. 124 (2021), 102912.
nanocomposites for environmental remediation applications, Ceram. Int. 46 (2020) [24] E. Clementi, D.-L. Raimondi, Atomic screening constants from SCF functions,
22171–22180, [Link] J. Chem. Phys. 38 (1963) 2686–2689.
[2] L. Ge, C. Han, J. Liu, Y. Li, Enhanced visible light photocatalytic activity of novel [25] J.C. Slater, Quantum Theory of Molecules and Solids, in: Symmetry and Energy
polymeric G-C3N4 loaded with Ag nanoparticles, Appl. Catal. Gen. 409–410 (2011) Bands in Crystals, vol. 2, 1965. New York.
215–222, [Link] [26] R. Sharma, P.K. Shishodia, A. Wakahara, R.M. Mehra, Investigations of highly
[3] K.M. Lee, C.W. Lai, K.S. Ngai, J.C. Juan, Recent developments of zinc oxide based conducting and transparent Sc doped ZnO films grown by the sol-gel process,
photocatalyst in water treatment technology: a review, Water Res. 88 (2016) Mater. Sci. 27 (2009).
428–448, [Link] [27] Z.Y. Zhong, T. Zhang, Microstructure and optoelectronic properties of titanium-
[4] B. Ai, X. Duan, H. Sun, X. Qiu, S. Wang, Metal-free graphene-carbon nitride hybrids doped ZnO thin films prepared by magnetron sputtering, Mater. Lett. 96 (2013)
for photodegradation of organic pollutants in water, Catal. Today 258 (2015) 237–239.
668–675, [Link] [28] C.-Y. Tsay, C.-W. Wu, C.-M. Lei, F.-S. Chen, C.-K. Lin, Microstructural and optical
[5] X. Li, D. Zhang, Z. Liu, Z. Li, C. Du, C. Dong, Materials science: share corrosion properties of Ga-doped ZnO semiconductor thin films prepared by sol–gel process,
data, Nature 527 (2015) 441–442, [Link] Thin Solid Films 519 (2010) 1516–1520.
[6] Y.-H. Chiu, T.-F.M. Chang, C.-Y. Chen, M. Sone, Y.-J. Hsu, Mechanistic insights into [29] S.R. Aghdaee, V. Soleimanian, B. Tayebi, Effect of Al doping on the
photodegradation of organic dyes using heterostructure photocatalysts, Catal 9 microstructural, optical and electrical properties of ZnO films, Superlattice.
(2019). Microst. 51 (2012) 149–162.
[7] H. Zhao, S. Chen, X. Quan, H. Yu, H. Zhao, Integration of microfiltration and [30] T. Miyata, S. Suzuki, M. Ishii, T. Minami, New transparent conducting thin films
visible-light-driven photocatalysis on g-C3N4 nanosheet/reduced graphene oxide using multicomponent oxides composed of ZnO and V2O5 prepared by magnetron
membrane for enhanced water treatment, Appl. Catal. B Environ. 194 (2016) sputtering, Thin Solid Films 411 (2002) 76–81.
134–140, [Link]

10
M.H. Saeed et al. Chemometrics and Intelligent Laboratory Systems 242 (2023) 104983

[31] K.A. Jeon, J.H. Kim, W.Y. Shim, W.Y. Lee, M.H. Jung, S.Y. Lee, Magnetic and [54] M. Lv, X. Xiu, Z. Pang, Y. Dai, L. Ye, C. Cheng, S. Han, Structural, electrical and
optical properties of Zn1− XMnxO thin films prepared by pulsed laser deposition, optical properties of zirconium-doped zinc oxide films prepared by radio frequency
J. Cryst. Growth 287 (2006) 66–69. magnetron sputtering, Thin Solid Films 516 (2008) 2017–2021.
[32] R. Ray, A. Kumar, A.K. Paul, S. Tyagi, Enhanced photocatalytic performance of [55] J.Z. Bloh, R. Dillert, D.W. Bahnemann, Ruthenium-modified zinc oxide, a highly
chromium doped zinc oxide nanoparticles, Trans. Indian Inst. Met. 69 (2016) active vis-photocatalyst: the nature and reactivity of photoactive centres, Phys.
1043–1048. Chem. Chem. Phys. 16 (2014) 5833–5845.
[33] P. Dhiman, K.M. Batoo, R.K. Kotnala, M. Singh, Fe-doped ZnO nanoparticles [56] L. Elsellami, H. Lachheb, A. Houas, Synthesis, characterization and photocatalytic
synthesised by solution combustion method, Micro & Nano Lett. 7 (2012) activity of Li-, Cd-, and La-doped TiO2, Mater. Sci. Semicond. Process. 36 (2015)
1333–1335. 103–114.
[34] X. Qiu, L. Li, G. Li, Nature of the abnormal band gap narrowing in highly crystalline [57] L.T.T. Nguyen, L.T.H. Nguyen, A.T.T. Duong, B.D. Nguyen, N. Quang Hai, V.
Zn 1-XCoxO nanorods, Appl. Phys. Lett. 88 (2006), 114103, [Link] H. Chu, T.D. Nguyen, L.G. Bach, Preparation, characterization and photocatalytic
10.1063/1.2185617. activity of La-doped zinc oxide nanoparticles, Materials 12 (2019) 1195.
[35] O. Altintas Yildirim, H. Arslan, S. Sönmezoğlu, Facile synthesis of cobalt-doped [58] S.Y. Bae, C.W. Na, J.H. Kang, J. Park, Comparative structure and optical properties
zinc oxide thin films for highly efficient visible light photocatalysts, Appl. Surf. Sci. of Ga-, in-, and Sn-doped ZnO nanowires synthesized via thermal evaporation,
390 (2016) 111–121, [Link] J. Phys. Chem. B 109 (2005) 2526–2531.
[36] D. Adler, H. Brooks, Theory of semiconductor-to-metal transitions, Phys. Rev. 155 [59] M. Ahmad, E. Ahmed, Z.L. Hong, Z. Iqbal, N.R. Khalid, T. Abbas, I. Ahmad, A.
(1967) 826–840, [Link] M. Elhissi, W. Ahmed, Structural, optical and photocatalytic properties of Hafnium
[37] M. Ferhat, A. Zaoui, R. Ahuja, Magnetism and band gap narrowing in Cu-doped doped zinc oxide nanophotocatalyst, Ceram. Int. 39 (2013) 8693–8700.
ZnO, Appl. Phys. Lett. 94 (2009), 142502. [60] Y. Cheng, L. Cao, G. He, G. Yao, X. Song, Z. Sun, Preparation, microstructure and
[38] R.B. Bylsma, W.M. Becker, J. Kossut, U. Debska, D. Yoder-Short, Dependence of photoelectrical properties of tantalum-doped zinc oxide transparent conducting
energy gap on x and T in Zn 1− x Mn x Se: the role of exchange interaction, Phys. films, J. Alloys Compd. 608 (2014) 85–89.
Rev. B 33 (1986) 8207. [61] J. Yang, X. Zhao, X. Shan, H. Fan, L. Yang, Y. Zhang, X. Li, Blue-shift of UV
[39] R. Krithiga, G. Chandrasekaran, Synthesis, strucutral and optical properties of emission in ZnO/graphene composites, J. Alloys Compd. 556 (2013) 1–5, https://
vanadium doped zinc oxide nanograins, J. Cryst. Growth 311 (2009) 4610–4614. [Link]/10.1016/[Link].2012.12.098.
[40] J. Singh, A. Rathi, M. Rawat, V. Kumar, K.-H. Kim, The effect of manganese doping [62] B.D. Ngom, T. Mpahane, N. Manyala, O. Nemraoui, U. Buttner, J.B. Kana, A.
on structural, optical, and photocatalytic activity of zinc oxide nanoparticles, Y. Fasasi, M. Maaza, A.C. Beye, Structural and optical properties of nano-structured
Composites, Part B 166 (2019) 361–370. tungsten-doped ZnO thin films grown by pulsed laser deposition, Appl. Surf. Sci.
[41] Y. Zhang, M.K. Ram, E.K. Stefanakos, D.Y. Goswami, Enhanced photocatalytic 255 (2009) 4153–4158.
activity of iron doped zinc oxide nanowires for water decontamination, Surf. [63] A.Y. Fasasi, M. Maaza, E.G. Rohwer, D. Knoessen, C. Theron, A. Leitch, U. Buttner,
Coating. Technol. 217 (2013) 119–123. Effect of Zn-doping on the structural and optical properties of BaTiO3 thin films
[42] V.C. Srivastava, Photocatalytic oxidation of dye bearing wastewater by iron doped grown by pulsed laser deposition, Thin Solid Films 516 (2008) 6226–6232.
zinc oxide, Ind. Eng. Chem. Res. 52 (2013) 17790–17799. [64] N. Babajani, S. Jamshidi, Investigation of photocatalytic malachite green
[43] M. Goswami, Enhancement of photocatalytic activity of synthesized Cobalt doped degradation by iridium doped zinc oxide nanoparticles: application of response
zinc oxide nanoparticles under visible light irradiation, Opt. Mater. 109 (2020), surface methodology, J. Alloys Compd. 782 (2019) 533–544.
110400. [65] M. Mujahid, Synthesis, Characterization and electrical properties of visible-light-
[44] X. Cai, Y. Cai, Y. Liu, H. Li, F. Zhang, Y. Wang, Structural and photocatalytic driven Pt-ZnO/CNT, Bull. Mater. Sci. 38 (2015) 995–1001.
properties of nickel-doped zinc oxide powders with variable dopant contents, [66] R.M. Mohamed, D. McKinney, M.W. Kadi, I.A. Mkhalid, W. Sigmund, Platinum/
J. Phys. Chem. Solid. 74 (2013) 1196–1203. zinc oxide nanoparticles: enhanced photocatalysts degrade malachite green dye
[45] M. Sajjad, I. Ullah, M.I. Khan, J. Khan, M.Y. Khan, M.T. Qureshi, Structural and under visible light conditions, Ceram. Int. 42 (2016) 9375–9381.
optical properties of pure and copper doped zinc oxide nanoparticles, Results Phys. [67] S. Haykin, Neural Networks: A Comprehensive Foundation, Prentice Hall PTR,
9 (2018) 1301–1309. 1998, 0132733501.
[46] R. Yogamalar, P.S. Venkateswaran, M.R. Benzigar, K. Ariga, A. Vinu, A.C. Bose, [68] D.W. Marquardt, An algorithm for least-squares estimation of nonlinear
Dopant induced bandgap narrowing in Y-doped zinc oxide nanostructures, parameters, J. Soc. Ind. Appl. Math. 11 (1963) 431–441.
J. Nanosci. Nanotechnol. 12 (2012) 75–83. [69] J. Jang, S. Anfis, Adaptive-network-based fuzzy inference system, IEEE Trans. Syst.
[47] J.-H. Lee, B.-O. Park, Characteristics of Al-doped ZnO thin films obtained by Man. Cybern. 23 (1993) 665–685.
ultrasonic spray pyrolysis: effects of Al doping and an annealing treatment, Mater. [70] Y. Zheng, H. Fang, H.O. Wang, Takagi-sugeno fuzzy-model-based Fault detection
Sci. Eng. B 106 (2004) 242–245, [Link] for networked control systems with markov delays, IEEE Trans. Syst. Man, Cybern.
[48] M.K. Satheesan, K.V. Baiju, V. Kumar, Influence of defects on the photocatalytic Part B 36 (2006) 924–929.
activity of niobium-doped ZnO nanoparticles, J. Mater. Sci. Mater. Electron. 28 [71] D. Karaboga, E. Kaya, Adaptive network based fuzzy inference system (ANFIS)
(2017) 4719–4724. training approaches: a comprehensive survey, Artif. Intell. Rev. 52 (2019)
[49] C. Wu, J. Shen, J. Ma, S. Wang, Z. Zhang, X. Yang, Electrical and optical properties 2263–2293.
of molybdenum-doped ZnO transparent conductive thin films prepared by dc [72] M.A. Mohammed, K.H. Abdulkareem, A.M. Dinar, B.G. Zapirain, Rise of deep
reactive magnetron sputtering, Semicond. Sci. Technol. 24 (2009), 125012. learning clinical applications and challenges in omics data: a systematic review,
[50] N. Güy, S. Çakar, M. Özacar, Comparison of palladium/zinc oxide photocatalysts Diagnostics 13 (4) (2023) 664.
prepared by different Palladium doping methods for Congo red degradation, [73] M.A. Mohammed, A. Lakhan, K.H. Abdulkareem, B. Garcia-Zapirain, A hybrid
J. Colloid Interface Sci. 466 (2016) 128–137. cancer prediction based on multi-omics data and reinforcement learning state
[51] T.N. Ravishankar, K. Manjunatha, T. Ramakrishnappa, G. Nagaraju, D. Kumar, action reward state action (SARSA), Comput. Biol. Med. 154 (2023), 106617.
S. Sarakar, B.S. Anandakumar, G.T. Chandrappa, V. Reddy, J. Dupont, Comparison [74] Yassine, S., & Stanulov, A. (2024). A comparative analysis of machine learning
of the photocatalytic degradation of trypan blue by undoped and silver-doped zinc algorithms for the purpose of predicting norwegian air passenger traffic. Int.J.
oxide nanoparticles, Mater. Sci. Semicond. Process. 26 (2014) 7–17. Math. Statist. Comput. Sci., 2, 28–43. [Link] .
[52] S. Lany, A. Zunger, Assessment of correction methods for the band-gap problem [75] Rajinikanth, V., Yassine, S., & Bukhari, S. A. (2024). Hand-Sketchs based
and for finite-size effects in supercell defect calculations: case studies for ZnO and Parkinson’s disease Screening using Lightweight Deep-Learning with Two-Fold
GaAs, Phys. Rev. B 78 (2008), 235104. Training and Fused Optimal Features. Int.J. Math. Statist. Comput. Sci., 2, 9–18.
[53] Y.J. Zhai, J.H. Li, X. Fang, X.Y. Chen, F. Fang, X.Y. Chu, Z.P. Wei, X.H. Wang, [Link] .
Preparation of cadmium-doped zinc oxide nanoflowers with enhanced
photocatalytic activity, Mater. Sci. Semicond. Process. 26 (2014) 225–230.

11