Artificial Olfactory System for Distinguishing Oil-Contaminated Soils
DINA SATYBALDINA, MARAT BAYDELDINOV, ALIYA ISSAINOVA, OLZHAS ALSEITOV,
ASSEM KONYRKHANOVA, ZHANAR AKHMETOVA, SHAKHMARAN SEILOV
Department of Information Security,
L.N. Gumilyov Eurasian National University,
2, Satpayev str., Astana, 010008,
REPUBLIC of KAZAKHSTAN
Abstract: - Oil-contaminated soils are a major environmental problem for Kazakhstan. Oil spills or leaks lead to
profound changes in the physical and agrochemical properties of the soil and the accumulation of hazardous
substances. Whilst there are many remote sensing techniques and complex laboratory methods for oil spill
detection, developing simple, reliable, and inexpensive tools for detecting the presence of pollutants in the soil
is a relevant research task. The study aims to research the possibilities of an electronic nose combining a
chemical sensor array with pattern recognition techniques to distinguish volatile organic compounds from
several types of hydrocarbon soil pollutants. An electronic nose system was assembled in our laboratory. It
includes eight gas metal oxide sensors, a humidity and temperature sensor, an analog-digital processing unit,
and a data communication unit. We measured changes in the electrical conductivity of sensors in the presence
of volatile organic compounds released from oil and petroleum products and samples of contaminated and
uncontaminated soils. The list of experimental samples includes six types of soils corresponding to different
soil zones of Kazakhstan, crude oil from three oil fields in Kazakhstan, and five types of locally produced fuel
oil (including gasoline, kerosene, diesel fuel, engine oil, and used engine oil). We used principal component
analysis to statistically process multidimensional sensor data, feature extraction, and collect the volatile
fingerprint dataset. Pattern recognition using machine learning algorithms made it possible to classify digital
fingerprints of samples with an average accuracy of about 92%. The study results show that electronic nose
sensors are sensitive to soil hydrocarbon content. The proposed approach based on machine olfaction is a fast,
accurate, and inexpensive method for detecting oil spills and leaks, and it can complement remote sensing
methods based on computer vision.
Key-Words: - artificial olfaction, crude oil, electronic nose, environment, machine learning, petroleum-derived
products, pollution, sensor, soil, volatile organic compounds.
Received: April 23, 2023. Revised: July 11, 2023. Accepted: September 8, 2023. Published: September 27, 2023.
1 Introduction
Oil spills are a global problem; natural, intentional,
or accidental oil spills can occur all around us.
Significant oil pollution of the environment occurs
in the territories adjacent to the sites of exploration,
development, and operation of hydrocarbon
deposits. Leaks from oil pipelines, spills when
pumping oil to sea vessels, or accidents during oil
transportation are also severe problems. One of the
most recent significant cases is an environmental
disaster on the Brazilian coast caused by the spill of
about 2.5 million tons of Venezuelan oil from a ship
that (intentionally or accidentally) dumped oil 700
km off the coast of Brazil, [1]. Finally, many
standard fuels are refined petroleum products, and
oil spills can occur during transportation, use, and
disposal in many places, including residential areas.
Oil and petroleum products pollution is
everywhere: in the soil layer, hydrosphere, and
atmosphere due to the high level of volatile organic
compounds emitted by spilled oil into the air, [2].
Oil introduces diverse chemical compounds into
soil, water, and air, disrupting the established
biogeochemical balance in ecosystems, [3].
Kazakhstan is the largest landlocked country in
Asia, and the country ranks 9th in the world in terms
of area. About 150 oil and 40 gas condensate fields
are being developed in Kazakhstan, [4]. Kazakhstan
has 30 billion barrels of proven oil reserves (12th
rank in the world), and our country has increased oil
production by 3.5 times over the past 30 years, [5].
At the same time, the development of the oil and
gas industry leads to contamination of soil, water
resources, and the atmosphere. Soil pollution in oil
production areas is becoming increasingly
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2023.19.89
Dina Satybaldina, Marat Baydeldinov,
Aliya Issainova, Olzhas Alseitov,
Assem Konyrkhanova, Zhanar Akhmetova,
Shakhmaran Seilov
E-ISSN: 2224-3496
951
Volume 19, 2023
significant, [6]. For example, 0.6 million hectares of
soil were detected as contaminated with oil in
western Kazakhstan, [7]. This region of the country
has the highest density of oil fields. Crude oil spills
disrupt soil structure and composition and reduce
plant nutrient availability and uptake, [3]. The soil
accumulates hydrocarbons and harmful
microelements. which inhibit the growth of plants
and microorganisms. The accumulation of aromatic
hydrocarbons in plant cells increases human
diseases, causing malignant tumors, [7].
Precise and rapid detection of crude oil and
petroleum-derived products (PDPs) are beneficial to
identifying the source and type of oil hydrocarbons,
accurately estimating oil spread areas, evaluating the
hazard level, and developing a response and
recovery treatment to reduce environmental effects.
Aircraft satellites' Remote Sensing (RS) tools
have been proven in oil spill detection and
monitoring, [8]. Optical and microwave sensors,
digital processing techniques, and pattern
recognition were used for image classification and
oil spill trajectory prediction, [9], [10]. The authors
of the work, [10], showed that the admixture of
nutrients impairs the accuracy of sensors when
recognizing images of oil spills. New approaches for
quickly and accurately classifying oil-contaminated
soils can use alternative ones that complement
remote sensing techniques based on computer
vision.
Crude oil contains certain hydrocarbons (HCs)
that have low boiling points and are classified as
volatile organic compounds (VOCs) with known
hazardous effects on human health and the air
ecosystem, [2].
This context has prompted researchers to use
artificial olfactory systems that mimic the
mammalian olfactory system (an electronic nose, e-
nose) and can discriminate odors by comparing their
smells to previously studied patterns, [11]. The
results of, [12], showed that the e-nose can
differentiate soil contamination due to gasoline and
diesel fuel leaks. The research in, [13], is another
example of the use of an artificial olfactory system.
The proposed approach made it possible to detect
petroleum products adsorbed on different surfaces.
In this work, we investigated whether e-noses
detect soil contamination and distinguish between
some types of soils, oil, and PDP pollutants In this
work, we investigated the performance of machine
olfactory systems for classifying soil types and soil
contaminants. A series of experiments were
conducted to measure the response of electronic nose
sensors to VOCs from all samples, including six
uncontaminated soil samples, crude oil from three
Kazakhstan oil fields, five PDPs from local
producers (gasoline, kerosene, diesel fuel, engine oil
and used engine oil), and soil samples with
introduced petroleum pollutant. Then, the principal
component analysis (PCA) was used for feature
extraction and collection of the volatile fingerprint
dataset. We evaluated the performance of decision
trees and k-nearest neighbor classifiers using
machine learning metrics. We demonstrated the high
sensitivity of the sensors and the selective
discrimination of VOC patterns of oil and PDP
samples depending on the type, nature, and relative
humidity of contaminated soils.
The rest of the paper consists of the following
sections. In section 2 we described the materials and
methods (materials, a description of the multisensory
e-nose system, datasets, research tools, and research
process). The experiment results on measuring
sensory responses to the presence of petroleum and
PDPs in soil samples, processing sensory data,
features extracting, and performance evaluating by
the machine learning algorithms are discussed in
Section 3. The conclusion and future research are
presented in Section 4.
2 Material and Methods
2.1 Samples
2.1.1 Soils
The relief forms of Kazakhstan are diverse
(highlands, forest-steppe, steppe, desert-steppe
cover, and desert), and it makes significant
differences between the types of soils found in
different geographical zones of the country.
Chernozems, meadow-chernozem soils, saline soils,
chestnut soils, and brown and grey-brown desert
soils represent the diversity of soil types in
Kazakhstan, [14].
Six soil types (chernozem, sand, birch grove soil,
clay, slaked lime, and peat) were selected for
laboratory experiments to cover a wide range of
commonly encountered soils in Kazakhstan. The
soil samples were collected locally from the surface
layers (at a depth of 5–20 cm) away from roads and
industries.
2.1.2 Contaminant
Crude oil and petroleum products spills release
VOCs that readily evaporate into the air and source
odors. The electronic nose can detect the effect of
even slight changes in the relative amount of
chemicals in the samples, [11]. Therefore, crude oil
from several fields and five types of PDPs produced
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2023.19.89
Dina Satybaldina, Marat Baydeldinov,
Aliya Issainova, Olzhas Alseitov,
Assem Konyrkhanova, Zhanar Akhmetova,
Shakhmaran Seilov
E-ISSN: 2224-3496
952
Volume 19, 2023
from different oil fractions with non-equal amounts
of VOCs were chosen as samples (Table 1).
Table 1. List of Samples
Petroleum
or PDP
sample type
Oil field\
Producing
company or
brand of sample
Sample’s
physicochemical
characteristics
Crude Oil
Alibekmola\
Kazakhoil Aktobe
LLP
Light (0.835 g/cm3),
sour (about 2%) oil
Crude Oil
Alibek Yuzhny \
Joint Stock
Company "Caspi
Neft TME"
Light (0.842 g/cm3),
sour (up to 1.33%) oil
Crude Oil
Kardasyn North \
Sagiz Petroleum
Company LLP
Heavy (0.945 g/cm3),
sweet (sour 0.2%) oil
Gasoline
Atyrau Oil
Refinery LLP \ A-
92-K5 with 92
octane number
Gasoline is produced
from the light fraction
of petroleum and
contains hydrocarbons
from four to twelve
carbon atoms per
molecule.
Diesel Fuel
Atyrau Refinery
LLP \ DT-A-K5
Altay-45
Diesel Fuel is produced
from the fraction of
crude oil next in density
to the gasoline fraction.
Diesel fuel, like
gasoline, contains a
significant amount of
volatile hydrocarbons
with a strong odor. It
contains hydrocarbons
from nine to twelve five
carbon atoms per
molecule.
Kerosene
PetroKazakhstan
Oil Products LLP \
RT
Kerosene is a mixture
of liquid hydrocarbons
containing eight to
fifteen carbons in the
hydrocarbon chain. Its
density is higher than
gasoline due to paraffin.
Engine Oil
High Industrial
Lubricants &
Liquids (HILL)
Corporation \
«HILL Universal»,
SAE 5W-30
Engine Oil is produced
from the dense
hydrocarbon fraction of
crude oil. It also
contains some additives.
Motor oils have a slight
odor because they
consist of non-light
hydrocarbons with more
than thirty carbon atoms
per molecule.
Used Engine
Oil
High Industrial
Lubricants &
Liquids (HILL)
Corporation \ Used
«HILL Universal»,
SAE 5W-30
Used Engine Oil may
change their
composition during use
and it may potentially
contain more harmful
contaminants than
unused engine oil.
Oil products used for mechanized land
cultivation are also a source of soil pollution, [15],
[16]. The study used two types of engine oil: a new
synthetic engine oil manufactured by High
Industrial Lubricants & Liquids (HILL) Corporation
and used engine oil (UMO) of the same brand
collected from a local garage.
2.2 E-nose
Our artificial olfactory system consists of a
multisensory array, an analog-to-digital conversion
unit, and a digital data transmission unit. MQ series
sensors from Hanwei Electronics were used as gas
sensors. Sensor color codes and target gases are
presented in the legends of Figure 1 (Appendix),
Figure 2 (Appendix) and Figure 3 (Appendix).
These sensors belong to the class of metal oxide
sensors.
The choice of this type of gas sensor is
associated with its low cost, small size, and
relatively low power consumption. The MOS
sensors use a two-electrode system in which the
sensitive layer of tin dioxide (SnO2) has a variable
resistance depending on the concentration and type
of gases or gas mixtures being studied. The
interaction of the target gas with the sensitive layer
occurs through a reversible redox reaction, as a
result of which the electrical properties of the SnO2
layer change. These electrical properties are
translated into measurable parameters such as sensor
resistances or voltage across a load resistor in a
sensor circuit.
MOS sensors, SHT75 temperature, and humidity
sensors were grouped on a separate board and
connected via an analog switch to an analog-to-
digital converter (ADC) on the Arduino Nano
module. The ADC has a 16-bit resolution and
converts input voltages from 0 to 5 volts into
integers from 0 to 65535 (216-1). The digitized data
is stored on a Raspberry Pi 4V microcomputer,
which has built-in support for 5G Wi-Fi and
Bluetooth 5. In the future, it is planned to implement
an intelligent sensor data processing unit on this
budget single-board computer.
MOS sensors need to be heated. Arduino cannot
provide enough power. Therefore, the system uses
an external power source connected via the USB
port. Sensors, a microcontroller, a microcomputer, a
power supply circuit, and a forced air supply unit
are placed inside a metal box.
2.3 Data Collection
Datasets were collected in a sequential series of
experiments measuring sensory responses from:
uncontaminated soil samples,
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2023.19.89
Dina Satybaldina, Marat Baydeldinov,
Aliya Issainova, Olzhas Alseitov,
Assem Konyrkhanova, Zhanar Akhmetova,
Shakhmaran Seilov
E-ISSN: 2224-3496
953
Volume 19, 2023
samples of Contaminant (crude oil and PDPs),
soil samples with introduced pollutants.
Soil samples of the same mass (70 grams), liquid
samples of oil and PDPs (volume 100 μl), and
mixtures of soils and pollutants were placed in a
glass vial connected to the nasal measuring
chamber.
Each experiment was repeated three times. The
output signal from the array of sensors, UADC, was
measured at a sampling frequency of 1 Hz during
the total time of the experiment for each sample (the
air was measured in the first part of the experiment,
then the sample was introduced, and the flow was
switched to air for 30 min in the final part).
Experimental data were recorded in separate files
(txt format). Each file consists of more than 7000
lines with ten attributes, namely data year-month-
day and time (hours-minutes-seconds), responses
from 8 sensors, temperature (degrees Celsius), and
humidity (%). The dataset contains all the data and
significant outliers.
The total number of records exceeded 170,000.
The data set was divided into two parts - 80% was
used as a training set and 20% of the data was used
at the classifier testing stage. In turn, part of the
training sample (30%) was used for validation to
control the training of machine learning models.
2.4 Data Analysis
Pattern recognition methods distinguish between
uncontaminated and contaminated soil samples, soil
samples, and petroleum products. Preliminary
feature extraction procedures are required to prepare
datasets for input to recognition systems.
We used principal component analysis to
statistically process measurements of sensory
responses to experimental samples and extract the
most significant features from multidimensional
sensory data.
Vector representations of the selected features as
volatile odor fingerprints from samples constituted
databases for training and testing classifiers based
on two machine learning algorithms (decision tree
and k-nearest neighbor method), [17].
We use code written in Python 3.9.15 and Scikit-
learn open source machine-learning package, [18].
The performance metrics (the confusion matrix,
accuracy, precision, and f-score) are used, [10].
For interactive calculations and experimental
data processing, a Python application has been
developed (in ipynb format) containing source
codes, input data, and calculation results in
numerical and graphical representation.
3 Results and Discussion
3.1 E-nose Results
of Figure 1 (Appendix), Figure 2 (Appendix) and
Figure 3 (Appendix) show examples of visualization
of experimental results in the form of time-
dependent electrical characteristics of sensors as
sensor responses to VOCs from the samples.
The figures show that the sensors of the gas analysis
system react differently to experimental samples of
uncontaminated and contaminated soils, crude oil,
and PDPs, which allows for obtaining volatile
fingerprints.
An analysis of the experimental data indicates
the presence of a correlation between the
physicochemical properties of crude oils from three
fields (Table 1) and the sensory responses to the
petroleum samples (Figure 1, Appendix). Light oils
contain more volatile hydrocarbons than denser oils.
Accordingly, the level of sensory responses to the
smell from VOCs is higher. The sulfur content in oil
can also affect the shape of sensory responses.
Gasoline corresponds to the lightest oil fraction.
This PDP has the wealthiest odor and fastest
evaporating. Therefore, we observe high-amplitude
sensory responses of the e-nose to the presence of
VOCs in motor gasoline and a rapid drop in sensory
response due to the high evaporation rate (Figure 2a,
Appendix). Diesel fuel corresponds to the average
fraction of oil, has fewer VOCs compared to
gasoline, and the sensory responses of the electronic
nose in the case of diesel fuel are less intense
(Figure 2b, Appendix), its shapes indicate the
average rate of evaporation and weathering.
Gasoline and kerosene have different chemical
chain lengths, but both have strong odors. Kerosene
contains paraffin and has a oilier structure.
Therefore, the experimental electrical characteristics
of MOS sensors in the presence of kerosene have a
high intensity, as in the case of gasoline, and
decrease more slowly (Figure 2c, Appendix).
Engine oils are thick mixtures of high molecular
weight hydrocarbons obtained by distillation of
heavy oil fractions (fuel oil and tar). The low
amplitudes of sensory responses to motor oil (Figure
2d, Appendix) indicate lower concentrations of
volatile organic compounds than lighter petroleum
products (gasoline, diesel fuel, or kerosene).
Remarkable results have been obtained with used
engine oil. Sensor responses indicate the presence of
a significant amount of VOCs in the used oil (Figure
2e, Appendix). The closeness of the used oil sensor
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2023.19.89
Dina Satybaldina, Marat Baydeldinov,
Aliya Issainova, Olzhas Alseitov,
Assem Konyrkhanova, Zhanar Akhmetova,
Shakhmaran Seilov
E-ISSN: 2224-3496
954
Volume 19, 2023
data to the gasoline and diesel fuel VOC fingerprints
indicates the possibility of:
- malfunctions in the engine that lead to the
ingress of gasoline or diesel fuel into the motor oil;
- or an increase in toxic components in oil
during use.
Figure 3 (Appendix) shows that the gas analysis
system reacts differently to experimental samples of
uncontaminated and contaminated soils, which
allows them to differentiate.
3.2 Data Analysis Results
The work uses eight gas sensors and a temperature
and humidity sensor. Sensory responses to the
samples with different quantities of the studied
substances and various temperature and air humidity
values represent multidimensional spaces of
dependent and independent features. Feature
extraction procedures are necessary for fast and
accurate classification of recognized objects, [17],
[19], [20].
We use the statistical method of principal
component analysis (PCA) to extract significant
features from sensory data and form a digital
fingerprint of each class of samples under study.
The first step of the method is data
standardization, [21]. The normalized value is the
ratio of the difference between the sensory response
value and the mean value to the standard deviation
value. Then we calculate the covariance matrix for
the normalized data, [22]. Calculating the
eigenvectors and eigenvalues allows us to select the
principal components. Finally, we can express the
standardized variables in terms of principal
component scores, [23].
An example of visualization of the results of
PCA analysis of sensory data is presented in Figure
4.
Fig. 4: Visualizing PCA results using Biplot for
sensor responses to the air and the crude oil from the
Alibekmola oil fields.
In Figure 4, each point represents sensor data
from two classes (air and oil). Let us recall that the
measurement of sensory responses for each sample
was carried out in stages: in the first part of the
experiment, the air was measured, then the sample
was introduced, and in the final part, the flow was
switched to stand for 30 min). It can be seen that
points belonging to the same class (oil or air) are
closer to each other, indicating that the features were
correctly extracted for constructing digital volatility
fingerprints.
Table 2 presents the performance estimates of
the volatile fingerprint classifiers of the studied
samples using two machine-learning algorithms.
Pattern recognition using machine learning
algorithms made it possible to classify digital
fingerprints of samples with an average accuracy of
about 92%.
Table 2 shows that the accuracy of digital
fingerprint recognition is above 90% in most cases.
The average classification accuracy of crude oil
brands and types of petroleum products is 92.15%
and 91.33% when using the decision tree and the
KNN method, respectively.
Table 2. Performance Results of Pattern
Recognition models to classify crude oils and PDPs
samples on test stage
Performance metrics
KNN
Decision Trees
Accuracy: 97.76 %
F1 SCORE: 97.72 %
PRECISION: 98.06 %
RECALL: 97.49 %
Accuracy: 98.32 %
F1 SCORE: 98.29 %
PRECISION: 98.53 %
RECALL: 98.11 %
Accuracy: 95.59 %
F1 SCORE: 95.53%
PRECISION: 96.2 %
RECALL: 95.25 %
Accuracy: 95.97 %
F1 SCORE: 95.92%
PRECISION: 96.51 %
RECALL: 95.65 %
Accuracy: 95.97 %
F1 SCORE: 95.92%
PRECISION: 96.51 %
RECALL: 95.65 %
Accuracy: 96.43 %
F1 SCORE: 96.23 %
PRECISION: 97.19%
RECALL: 95.55%
Accuracy: 96.23 %
F1 SCORE: 96.19%
PRECISION: 96.71%
RECALL: 95.93 %
Accuracy: 97.32 %
F1 SCORE: 97.29 %
PRECISION: 97.53 %
RECALL: 97.11 %
Accuracy: 85.69 %
F1 SCORE: 85.47 %
PRECISION: 85.64 %
RECALL: 85.35 %
Accuracy: 87.8 %
F1 SCORE: 87.57 %
PRECISION: 87.9 %
RECALL: 87.37 %
Accuracy: 95.66 %
F1 SCORE: 95.6 %
PRECISION: 96.25 %
RECALL: 95.31 %
Accuracy: 96.22 %
F1 SCORE: 96.18 %
PRECISION: 96.71 %
RECALL: 95.93 %
Accuracy: 72.45 %
F1 SCORE: 72.1 %
PRECISION: 76.31 %
RECALL: 80.15 %
Accuracy: 72.96 %
F1 SCORE: 72.58 %
PRECISION: 76.54 %
RECALL: 80.52 %
The decrease in classification accuracy for diesel
fuel and used engineering oil is due to the proximity
of sensory responses for these samples. Therefore,
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2023.19.89
Dina Satybaldina, Marat Baydeldinov,
Aliya Issainova, Olzhas Alseitov,
Assem Konyrkhanova, Zhanar Akhmetova,
Shakhmaran Seilov
E-ISSN: 2224-3496
955
Volume 19, 2023
part of the data for diesel fuel supplied to the input
of the classifiers is incorrectly recognized as
volatility fingerprints of used motor oil, and vice
versa.
4 Conclusion
In the present paper, we investigated artificially
polluted soils to detect soil contamination and
distinguish between some types of soil, oil, and PDP
pollutants using alternative sensor technologies that
can replace or supplement computer vision and RS
techniques.
We examined six soil types found in different
geographic areas of Kazakhstan, eight different
pollutants, including crude oil from three
Kazakhstan fields, and commercial gasoline,
kerosene, diesel fuel, motor oil, and used motor oil.
The sensor responses to volatile organic compounds
of soil, crude oil, and petroleum product samples
recorded by the electronic nose were processed
using a statistical analysis method. After the feature
extraction stage, feature vectors were used to train
and test classifiers based on a machine learning
algorithm. The experimental results showed that the
artificial olfactory system is sensitive to different
types of soil and the composition of petroleum
products. A machine learning model implemented in
Python recognizes contaminated and
uncontaminated soils with high accuracy and the
kind of oil and petroleum products. The proposed
approach to detecting oil-contaminated soils based
on inexpensive and compact devices such as an
electronic nose is a good alternative to oil spills'
current remote sensing methods.
The proposed approach will be used in future
studies to determine the source and type of oil
pollution on soil samples from oil production fields
and other contaminated areas.
References:
[1] Marcelo de Oliveira Soares et al., Oil spill in
South Atlantic (Brazil): Environmental and
governmental disaster, Marine Policy, Vol.
115, 2020, 103879.
[2] Ana Cláudia Souza et al., Oil spills
characteristics, detection, and recovery
methods: A systematic risk-based view,
Journal of Loss Prevention in the Process
Industries, Vol. 80, 2022, 104912.
[3] Hussein, Z. S., Hamido, N., Hegazy, A. K., El-
Dessouky, M. A., Mohamed, N. H., & Safwat,
G. (2022). Phytoremediation of crude
petroleum oil pollution: a review, Egypt. J.
Bot., Vol. 62, No.3, 2022, pp.611-640.
[4] Bealessio B., Alonso N., Mendes N., Sande A.,
& Hascakir B., A review of enhanced oil
recovery (EOR) methods applied in
Kazakhstan, Petroleum, Vol. 7 (1), 2021, pp.1–
9.
[5] BP Statistical Review of World Energy,
accessed 8 July 2021, 70th Edition
https://www.bp.com/content/dam/bp/business-
sites/en/global/corporate/pdfs/energy-
economics/statistical-review/bp-stats-review-
2021-full-report.pdf.
[6] Alimbaev, T., Yermagambetova, K.,
Kabyltayeva, S., Issayev, A., Kairat, Z., &
Mazhitova, Z., Environmental problems of the
oil and gas industry in Kazakhstan, E3S Web of
Conferences, Vol. 215, 2020, p.03008.
[7] Akhmetov L.I. et al., Recent Advances in
creating biopreparations to fight oil spills in
soil ecosystems in the sharply continental
climate of the Republic of Kazakhstan,
Processes, Vol.10, No.3, 2022, 549.
[8] Fingas M., Brown C., & Daniels J., An update
on oil spill remote sensing, Advances in
Environmental Research, Vol.80, 2021, pp.95-
137.
[9] Shamsudeen T., Advances in remote sensing
technology, machine learning and deep
learning for marine oil spill detection,
prediction and vulnerability assessment,
Remote Sensing, Vol. 12, No. 20, 2020, p.3416.
[10] Al-Ruzouq R. et al., Sensors, features, and
machine learning for oil spill detection and
monitoring: A review, Remote Sensing, Vol.12,
No.20, 2020, p.3338.
[11] Zaytsev V. et al., Rapid and accurate quality
assessment method of recycled food plastics
VOCs by electronic nose based on Al-doped
zinc oxide, Journal of Cleaner Production,
Vol. 418, 2023, 138042.
[12] Bieganowski A. et al., Evaluation of
hydrocarbon soil pollution using e-nose,
Sensors, Vol.18, No.8, 2018, 2463.
[13] Ferreiro-González, M., Barbero, G. F., Palma,
M., Ayuso, J., Álvarez, J. A., & Barroso, C. G.,
Characterization and differentiation of
petroleum-derived products by E-nose
fingerprints, Sensors, Vol.17, No.11, 2017,
2544.
[14] Pachikin, K., Erokhina, O., & Funakawa, S.,
Soils of Kazakhstan, their distribution and
mapping, in Novel measurement and
assessment tools for monitoring and
management of land and water resources in
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2023.19.89
Dina Satybaldina, Marat Baydeldinov,
Aliya Issainova, Olzhas Alseitov,
Assem Konyrkhanova, Zhanar Akhmetova,
Shakhmaran Seilov
E-ISSN: 2224-3496
956
Volume 19, 2023
agricultural landscapes of Central Asia,
Springer International Publishing, Switzerland,
2014, pp.519-533.
[15] Vazquez-Duhalt R., Environmental impact of
used motor oil, Science of the total
environment, Vol.79.1,1989, pp.1–23.
[16] Mishra A., Siddiqi H., Kumari U., Behera I.,
Mukherjee S. & Meikap B., Pyrolysis of waste
lubricating oil/waste motor oil to generate high-
grade fuel oil: A comprehensive review,
Renewable and Sustainable Energy Reviews,
Vol. 150, 2021, 111446.
[17] Isah Salim Ahmad, Shuai Zhang, Sani Saminu,
Lingyue Wang, Abd El Kader Isselmou,
Ziliang Cai, Imran Javaid, Souha Kamhi,
Ummay Kulsum, Deep Learning Based on
CNN for Emotion Recognition Using EEG
Signal, WSEAS Transactions on Signal
Processing, vol. 17, 2021, pp.28-40.
[18] Hao J., & Ho T., Machine learning made easy:
a review of a scikit-learn package in a python
programming language, Journal of Educational
and Behavioral Statistics, vol.44(3), pp.348-
361, March 2019.
[19] Md. Salah Uddin Yusuf, Azmol Ahmed Fuad,
Real Time Implementation of Face Recognition
based Smart Attendance System, WSEAS
Transactions on Signal Processing, vol. 17,
2021, pp.46-56.
[20] Meili Dai, Intelligent Correction System of
Students’ English Pronunciation Errors Based
on Speech Recognition Technology, WSEAS
Transactions on Advances in Engineering
Education, vol. 18, 2021, pp.192-198.
[21] Jackson J. E., A user's guide to principal
components, Toronto, Canada: John Wiley &
Sons Inc., 2005.
[22] Fadda ML, Moussaoui A. Faults Detection and
Diagnosis Approach Using PCA and SOM
Algorithm in PMSG-WT System, WSEAS
Transactions on Systems, Vol. 16, 2017,
pp.159-67.
[23] Radhia Fazai, Okba Taouali, Nasreddine
Bouguila, Adaptive Fault Detection Methods
Based on PCA Technique, WSEAS
Transactions on Circuits and Systems, vol. 14,
2015, pp.337-346.
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2023.19.89
Dina Satybaldina, Marat Baydeldinov,
Aliya Issainova, Olzhas Alseitov,
Assem Konyrkhanova, Zhanar Akhmetova,
Shakhmaran Seilov
E-ISSN: 2224-3496
957
Volume 19, 2023
APPENDIX
Fig. 1: The time-dependent electrical characteristics of e-nose sensors for crude oil from three Kazakhstan oil
fields: (a) Alibekmola, (b) Alibek Yuzhny, (c) Kardasyn North.
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2023.19.89
Dina Satybaldina, Marat Baydeldinov,
Aliya Issainova, Olzhas Alseitov,
Assem Konyrkhanova, Zhanar Akhmetova,
Shakhmaran Seilov
E-ISSN: 2224-3496
958
Volume 19, 2023
Fig. 2: The time-dependent electrical characteristics of e-nose sensors for PDPs: (a) Gasoline; (b) Diesel fuel;
(c) Kerosene; d) Engine oil; (e) Used engine oil.
Fig. 3: The time-dependent electrical characteristics of e-nose sensors for the uncontaminated and contaminated soil
samples: (a) chernozem (dry and wet), (b) sand (dry and wet), (c) birch grove soil (dry and wet), d) chernozem
sand + kerosene; (e) sand + Kerosene, (f) birch grove soil + kerosene.
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2023.19.89
Dina Satybaldina, Marat Baydeldinov,
Aliya Issainova, Olzhas Alseitov,
Assem Konyrkhanova, Zhanar Akhmetova,
Shakhmaran Seilov
E-ISSN: 2224-3496
959
Volume 19, 2023
Contribution of Individual Authors to the
Creation of a Scientific Article (Ghostwriting
Policy)
- Dina Satybaldina: Supervision, Conceptualization,
Methodology, Validation, Investigation, Writing –
original draft.
- Marat Baydeldinov: Experimental Methodology,
Investigation, Writing – original draft.
- Aliya Issainova, PhD student: Experiments
execution, Data curation, ML models
implementation in Python, Data analysis.
- Olzhas Alseitov, Master student: Experiments
execution, Data curation.
- Assem Konyrkhanova: Methodology, Data
Analysis, Writing – original draft.
- Zhanar Akhmetova: Methodology, Investigation,
Validation.
- Shakhmaran Seilov: Project supervising,
Resources.
Sources of Funding for Research Presented in a
Scientific Article or Scientific Article Itself
This research is funded by the Science Committee
of the Ministry of Science and Higher Education of
the Republic of Kazakhstan (Grant No.
AP14872171).
Conflict of Interest
The authors have no conflict of interest to declare.
Creative Commons Attribution License 4.0
(Attribution 4.0 International, CC BY 4.0)
This article is published under the terms of the
Creative Commons Attribution License 4.0
https://creativecommons.org/licenses/by/4.0/deed.en
_US
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2023.19.89
Dina Satybaldina, Marat Baydeldinov,
Aliya Issainova, Olzhas Alseitov,
Assem Konyrkhanova, Zhanar Akhmetova,
Shakhmaran Seilov
E-ISSN: 2224-3496
960
Volume 19, 2023