Classifying PDO Kalamata Olive Oil from Geographic Origins of the
Messenia Region based on Statistical Machine Learning
THEODOROS ANAGNOSTOPOULOS1, IOAKEIM SPILIOPOULOS2
1Department of Business Administration,
University of West Attica,
12241 Athens,
GREECE
2Department of Food Science and Technology,
University of Peloponnese,
24100 Kalamata,
GREECE
Abstract: - Kalamata is a smart city located in southeastern Greece in the Mediterranean basin and it is the
capital of the Messenia regional unit. It is known for the famous Protected Designation of Origin (PDO)
Kalamata olive oil produced mainly from the Koroneiki olive variety. The PDO Kalamata olive oil, established
by Council regulation (EC) No 510/2006, owes its quality and special characteristics to the geographical
environment, olive tree variety, and human factor. The PDO Kalamata olive oil is produced exclusively in the
regional unit of Messenia, being the main profit of local farmers. However, soil chemical composition,
microclimates, and agronomic factors are changed within the Messenia spatial area leading to differentiation of
PDO Kalamata olive oil characteristic. In this paper, we use statistical machine learning algorithms to
determine the geographical origin of Kalamata olive oil at PDO level based on synchronous
excitation−emission fluorescence spectroscopy of olive oils. Evaluations of the statistical models are promising
for differentiating the origin of PDO Kalamata olive oil with high values of prediction accuracy thus enabling
companies that process and bottle kalamata olive oil to choose olive oil from a specific region of Messenia that
fulfills certain characteristics. Concretely, the current research effort focuses on a specific olive oil variety
within a limited geographic region. Intuitively, future research should also focus on validation of the proposed
methodology to other olive oil varieties and production areas.
Key-Words: - PDO Kalamata olive oil, synchronous emission-excitation, fluorescence spectroscopy, statistical
machine learning, data fusion, data visualization, multiclass classification, model evaluation.
1 Introduction
Smart agriculture is the dimension of the smart city
concept aiming to define methods of efficient
geographic cultivation in rural areas, [1]. The
cropping of plants useful for cities’ citizens is the
main area of interest for smart farming, [2].
Specifically, in the area of olive oil farmers in the
Messenia region of Greece produce the protected
designation of origin (PDO) extra virgin olive oil
with the name Kalamata olive oil in the rural areas
of the smart city Kalamata the Koroneiki olive
variety is almost exclusively cultivated which
produces the extra virgin olive oil with organoleptic
properties, [3], [4]. Such areas provide farmers the
capability to gain more income since specific olive
oil microclimates affect the quality of the selected
variety, [5].
To protect olive oil quality and prevent its
adulteration, global governmental agencies like the
European Commission, International Olive Council,
Codex Alimentarius, etc have developed standards
to regulate olive oil by establishing a set of physical,
chemical, and organoleptic characteristics, [6]. The
traditional chemical methods to ensure olive oil
quality are focused on the identification and
quantification of pre-defined compounds or classes
of compounds of olive oil according to the
regulations of the above-mentioned global
governmental agencies. These methods are time-
consuming and demand expensive apparatus. The
same for the detection of olive oil adulteration
Received: June 9, 2023. Revised: February 24, 2024. Accepted: April 3, 2024. Published: May 13, 2024.
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2024.20.15
Theodoros Anagnostopoulos, Ioakeim Spiliopoulos
E-ISSN: 2224-3496
137
Volume 20, 2024
although these methods fail to detect the
adulteration from certain adulterants.
In recent years the non-targeted analysis has
attracted much attention. This approach focuses on
screening the olive oil without any prior knowledge
of chemical composition. In this approach, we used
analytical techniques that produce a signal which is
affected by all the compounds (i.e., metabolites)
present in olive oil. These methods shorten the
analysis process but a vast number of data sources
are required to perform data analytics based on
statistical machine learning algorithms, [7].
To assess the quality of gathered olive oil there is
a need to incorporate specific Internet of Things
(IoT) devices, [8]. A device that is commonly used
for such a process is fluorescence spectroscopy,
which is calibrated accordingly to perform
differences of excitation and emission radiation to
the olive oil sample, [9]. Concretely, fluorescence
spectrometry has been used extensively in the past
years due to its efficient precision in recognizing
chemical components of olive oil samples thus
exploiting its overall quality, [10]. Specifically,
adopted technology can access input data from olive
oil sample sources to measure optimally the
chemical ingredients of a given olive oil sample as
well as to be able to discriminate the olive oil
quality categories as well as its origin, [11].
Intuitively, fluorescence spectra technology can
detect in high effectiveness adulteration of olive oil
with other lower-quality oils, such as sunflower oil
or soybean oil, [12]. Collecting samples from
different geographical origins enables the generation
of different data sources, [13]. Exploited data can be
visualized and analyzed by statistical machine
learning algorithms. Intuitively, the application of
statistical classifiers enables the classification of
olive oil samples into certain categories able to
differentiate the quality of each sample, [14].
In this paper, we input synchronous emission-
excitation fluorescence spectra of PDO Kalamata
olive oils of different local geographic origins from
Messenia to observe the resulting data sources. DPO
Kalamata olive oils were from the areas of (1) Aris,
(2) Thouria, (3) Verga, (4) Arfara, and (5)
Meligalas. Subsequently, we input such data sets to
certain statistical machine learning algorithms to
assess which of them has optimal results to
recognize the different local cultivations. Adopted
statistical learning algorithms are evaluated with
certain evaluation methods and metrics to observe
an optimal classification of input olive oil samples.
The outcome of the research effort is to be able to
characterize the specific origin of each PDO
Kalamata olive oil (within the Messenia region) thus
companies that process and bottlebottle Kalamata
olive oil can choose olive oil from specific regions
of Messenia that fulfils superior characteristics.
The rest of the paper is organized as follows. In
Section 2 it is presented the prior work in the
research effort area. Section 3 defines the adopted
data model. In Section 4 evaluation parameters are
defined. In Section 5 experiments are performed and
results are observed. Section 6 discusses the
strengths and the weaknesses of the proposed
research effort, while Section 7 concludes the paper
and proposes future work.
2 Prior Work
Extra virgin and virgin olive oil have recently
attracted consumer interest because of their quality,
and its potential health benefits derived from their
consumption. The high price of extra virgin olive oil
and its reputation makes olive oil a target for
fraudsters. Significant research has been performed
in the literature in the area of olive oils’ analysis,
classification, authentication, origin, and
adulteration. Spectroscopic techniques such as
ultraviolet-visible (i.e., UV–Vis) absorption [15],
[16], fluorescence spectroscopy [17], Raman
spectroscopy [18], mass spectrometry [19], nuclear
magnetic resonance [20] and FT-NIR [21] have
been proposed to classify and detect adulteration
and origin of olive oil. Classification based on
statistical machine learning is used to compare
virgin olive oil quality in [22]. Fluorescence
spectroscopy is used along with principal
component technology and factorial discriminant
analysis for monitoring and classifying certain
virgin olive oil varieties. Raman spectroscopy is
incorporated in [23], to identify olive oil quality
using classification techniques. Intuitively, the
adopted method used a one-dimensional
convolutional deep-learning neural network to
observe optimal classification results. Portable
Raman spectroscopy is used in [24], to provide
quality assessment and control of several olive oil
varieties. Subsequently, the proposed method
adequately covers the cases of adulterated
compound low-quality oils within the virgin olive
oil.
Classification and authentication techniques are
incorporated in [25], to distinguish the origins of
virgin olive oil. Specifically, it is proposed an
authentication process is proposed to analyze
volatile olive oil compounds and chemometrics to
assess the quality of certain olive oil varieties within
a local geographic area. Statistical machine learning
algorithms are incorporated in [26], to classify
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2024.20.15
Theodoros Anagnostopoulos, Ioakeim Spiliopoulos
E-ISSN: 2224-3496
138
Volume 20, 2024
specific olive oil varieties. Concretely, the adopted
method uses discrimination techniques to input
machine learning algorithms with spectroscopic data
thus achieving effective prediction accuracy of olive
oil behavior by exploiting fusion emission and
absorption. Fluorescence spectroscopy is
incorporated in, [27], to classify the high quality of
olive oil. Intuitively, the proposed method assesses a
certain thermal oxidation technique, which exploits
the potentiality of an Ultra Violet (UV) fluorescence
spectroscopy system to perform specific imaging
classification of extra virgin olive oil varieties.
A time series classification algorithm is
incorporated in [28], which can distinguish several
virgin olive oil varieties. Subsequently, a statistical
transformation of the generated input data sources is
performed on each virgin olive oil variety to assess
the ensemble classification schema thus observing
optimal values of the prediction accuracy evaluation
metric. A multivariate classification analysis is
incorporated in [29], which can distinguish extra
virgin olive oils. Concretely, the adopted method is
based on Fourier Transform Infrared Spectroscopy
(FTIR) along with multivariate analysis to classify
virgin olive oils’ geographic origins, which come
from several producing countries. Adulterated olive
oil, in [30], can be discriminated with the
incorporation of Attenuated Total Reflection (ATR)
and FTIR spectroscopy technologies. Intuitively, the
proposed methods are capable of distinguishing pure
samples of virgin olive oil from different oil blends
by exploiting the potentiality of partial least squares
discriminant analysis (PLS-DA) applied to given
olive oil compounds.
Methods and applications for distinguishing
several extra virgin olive oils’ local geographic
origins are proposed in the literature, [31].
Specifically, the classification of olive oil
geographic origins is based on certain chemometric
data sources. Such chemometric data are generated
from several olive oils compounds, which input the
fluorescence spectroscopy decision-making models
to achieve optimal prediction accuracy.
Synchronous scanning of chemometric data sources
produced by significantly detailed fluorescence
spectroscopy measurements is also supported in
certain research efforts, [32]. Such knowledge is
then exploited by specific statistical classification
learning models, which can distinguish several
varieties of edible extra virgin olive oils. Edible
olive oils' premium quality is assessed in the
literature, [33]. Concretely, such ability is achieved
by the incorporation of synchronous fluorescence
spectroscopy, which can differentiate the
quantification of tocopherols from the input olive oil
compounds.
Geographic origins of olive oil varieties, [34],
are feasible due to the incorporation of chemometric
analysis. Specifically, such advanced analytical
methodology, which is applied to data sources can
predict olive oil’s registered designation with
optimal precision taking into consideration
synchronous excitation and emission of
fluorescence spectra values. Rapid spectroscopic
methods (Vis–NIR and FT-MIR) along with PLS
analysis were applied to study thermal stress of
virgin olive oils, [35]. Concretely, due to the
manipulation of generated data sources to certain
statistical learning models, which can evaluate
optimally spectroscopic and chemometric
technologies. Pattern recognition is also
incorporated in extra virgin olive oil varieties
classification, [36]. Intuitively, near-infrared
spectrometry provides the technical methodology to
assess the strengths of screening methods, which are
then used to authenticate extra virgin olive oils from
near local geographic origins. Shelf-Life olive oil
varieties are monitored and then classified to certain
geographic origins, [37]. Subsequently, IoT sensors
and actuators technology is exploited to enhance
fluorescence spectroscopy characteristics thus being
able to correctly assess the multiclass classification
process, which is based on certain statistical
learning models.
There are many research approaches that deal
with the origins of olive oil based on statistical
machine learning. Promising efforts incorporate
generated data from several chemometric
technologies. However, data manipulation requires
improvement to distinguish data interconnections,
which can provide efficient results. In this research
effort, fluorescence spectroscopy is exploited by
applying enhanced data preprocessing. Such
optimized data sources are then used by a statistical
machine learning algorithm to perform multiclass
classification to distinguish between certain local
cultivations’ origins of the Koroneiki olive oil
variety in the smart city of Kalamata, which is
located in the Messenia region, Greece.
3 Data Model
Data provided to perform analytics are synchronous
emission-excitation fluorescence spectra. These
spectra were recorded on a Perkin Elmer LS55
spectrofluorometer using solution 1% w/v olive oil
in n-hexane, where Δλ (i.e., the difference between
excitation and emission wavelength) was adjusted to
30 nm, [38]. The excitation and emission slit were
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2024.20.15
Theodoros Anagnostopoulos, Ioakeim Spiliopoulos
E-ISSN: 2224-3496
139
Volume 20, 2024
tuned to 4 nm. The scan rate was 50nm/min. Each
olive oil sample was measured triplicate using the
new freshly prepared solution. Each measurement of
an olive oil sample was statistically handled as a
different sample of the same origin.
Such data have a certain structure. Specifically,
observed data sources are collected from PDO
Kalamata olive oil produced in a variety of local
areas in the rural areas of the smart city of Kalamata
in the Messenia region, Greece. Concretely, there
are collected data from 29 olive oil samples from
the local cultivation areas of (1) Aris, (2) Thouria,
(3) Verga, (4) Arfara, and (5) Meligalas. Intuitively,
local cultivation areas are classified into the
following 5 classes, namely: (1) Aris: Class 1, (2)
Thouria; Class 2, (3) Verga: Class 3, (4) Arfara:
Class 4, and (5) Meligalas: Class 5. Subsequently,
the distribution of collected data samples per class
are as follows: (1) 2 samples from Aris, (2) 2
samples from Thouria, (3) 7 samples from Verga,
(4) 15 samples from Arfara, and (5) 3 samples from
Meligalas.
3.1 Data Structure
Synchronous emission-excitation fluorescence
spectra are composed of two-dimensional
coordinates, i.e., , where is the
identifier of each olive oil class, where: refers
to Aris, refers to Thouria , refers to
Verga, refers to Arfara, and refers to
Meligalas local geographic origins. Concretely,
dimension depicts the emission wavelength
measured in nanometers (i.e., ) for all sample
classes while dimension depicts
photoluminescence intensity, which is an arbitrary
net number based on internal calibration of the
spectrofluorometer device for each of the data
sample classes.
3.1.1 Visualizing Initial Data Samples
Intuitively, according to the initial data sample
measurements, assigned specific values for
dimension in the initial interval such as:
for all data sample classes (i.e., the 5
classes of local cultivation origins). Subsequently,
initial data samples are averaged according to
values based on each olive oil class.
Such average is performed to be able to provide a
simple and easily understandable visualization
based on initial data samples for each of the 5
classes. Concretely, averaged values in dimension
is varying according to the examined olive oil data
sample classes, as follows: (1) in case of Aris has
initial values in the interval, ,
(2) in case of Thouria has initial values in the
interval, , (3) in case of Verga
has initial values in the interval,
, (4) in case of Arfara has initial
values in the interval, , and (5)
in case of Meligalas has initial values in the
interval, . Figure 1, visualizes
the initial data samples (i.e., synchronous
photoluminescence spectra) per certain class of
olive oil geographic origin. It can be observed that
classes are not easily distinguished from each other
based on initial data measurements. This should be
treated accordingly with the data fusion process to
have a clearer view of how classes could be more
easily distinguished.
Fig. 1: Initial data (i.e., synchronous emission-
excitation fluorescence spectra) are assigned to a
specific class of certain local origin of PDO
Kalamata olive oil. Observed spectra were recorded
at Δλ = 30
3.2 Data Fusion
Data fusion is a widely adopted method used in
machine learning literature in case there is a need to
understand in depth the inherent complexity of
initial data sources. Concretely, the data fusion
process is applied to experimental input data sources
to visualize the statistical qualitative trends of the
data provided, thus being able to incorporate
efficient machine learning algorithms to observe
optimal results. Intuitively, initial data samples are
transformed according to a specific data fusion
process to remove outliers and missing values that
occurred during the initial measurement process
performed by the fluorescence spectroscopy device.
Specifically, such a data fusion process can provide
easily distinguished classes between each other in
contrast to the initial data due to the adopted
transformation. Intuitively, dimension values of
each data sample are transformed into interval
values. Such intervals form the predictive attributes,
which will input the statistical learning classifier to
predict the correct class of olive oil origin.
Concretely, is the assigned identifier of
each transformed wavelength interval (i.e.,
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2024.20.15
Theodoros Anagnostopoulos, Ioakeim Spiliopoulos
E-ISSN: 2224-3496
140
Volume 20, 2024
predictive attribute) of the olive oil components.
Subsequently, for it holds that in
case of: refers to that is
assigned to the predictive attribute of ‘tocopherols’,
refers to that is assigned to
the predictive attribute of ‘phenolic compounds’,
refers to that is assigned to
the predictive attribute of ‘oxidation products of
triglycerides’, refers to that
is assigned to the predictive attribute of ‘oxidation
products of tocopherols’, and refers to
that is assigned to the defined predictive
attribute of ‘chlorophylls’, components.
Subsequently, measured values in dimension
have an arbitrary initial distribution according to the
examined olive oil data sample class as produced by
the synchronous photoluminescence spectra. Such
values are transformed into
aggregated values for
each olive oil class origin, , and each assigned
identifier, to each transformed wavelength
interval, , (i.e., a certain predictive attribute)
according to specific olive oil measured compounds
of local geographic origin. Concretely, it holds that
is a transformed average value that is assigned to
each fused predictive attribute (i.e., ) of a certain
data sample class. There are specific
fused values
given certain data instances of the initial data
observed by the 29 olive oil samples (i.e., class
values) from different local reas for each of the
interval values, (i.e., predictive attributes).
Fig. 2: Fused data (i.e., based on initial synchronous
emission-excitation fluorescence spectra) are
assigned to a specific class of certain local origin of
PDO Kalamata olive oil. Observed spectra were
recorded at Δλ = 30
3.2.1 Visualizing Fused Data Samples
Intuitively, according to the fused data sample
measurements, for certain , is assigned
specific values for the interval values (i.e.,
predictive attributes) according to for the
fused data sample classes (i.e., the 5 classes of
origins). Subsequently, fused data samples are
averaged according to
values for
each olive oil class. Such average is performed to be
able to provide a simple and easily understandable
visualization based on fused data samples for each
of the 5 classes. Concretely, averaged fused values
in
(i.e., predictive attributes’ value range) is
varying according to the examined olive oil data
sample classes, as follows: (1) in case of Aris
has observed fused data values of:
, (2) in case
of Thouria,
has certain fused
values,
, (3)
in case of Verga
has fused
values,
, (4)
in case of Arfara
has fused
values,
, and
(5) in case of Meligalas
has
fused values,
,
Figure 2, visualizes the fused data samples per
certain class of olive oil origin. It can be observed
that classes are now more easily distinguished from
each other based on fused data measurements. This
is the reason for treating initial data samples with a
data fusion process to have a clearer view of how
classes could be more easily distinguished.
4 Evaluation Parameters
Assessing the performance of the adopted statistical
machine learning algorithm, certain valuation
methods and evaluation metrics should be
incorporated to perform specific experiments and
observe derived results.
4.1 Evaluation Method
To evaluate a statistical machine learning algorithm
there are used certain evaluation methods. Authors
adopt one of the widely used evaluation methods,
due to its simplicity and optimum results, which is
10-fold cross-validation, [39]. Specifically, such an
evaluation method divides the input dataset into 10
equal sized parts and then in a certain loop
incorporates the first 9 parts to train the statistical
learning classification algorithm and the remaining
1 to test the classifier. This process is repeated until
all the parts are used for training and testing. The
proposed evaluation method is adopted in the
machine learning methodology since it provides
effective results based on certain input data able to
explain the observed data source’s predictive
analytics behavior.
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2024.20.15
Theodoros Anagnostopoulos, Ioakeim Spiliopoulos
E-ISSN: 2224-3496
141
Volume 20, 2024
Table 1. Confusion matrix
4.2 Evaluation Metrics
Given the evaluation method, which is proposed to
support the experimental setup there is a need to
adopt specific evaluation metrics. Such metrics are:
(1) prediction accuracy, (2) correctly classified
instances, and (3) confusion matrix that can assess
the efficiency of a statistical classification
algorithm.
4.2.1 Prediction Accuracy
The effectiveness of the adopted statistical learning
algorithm is assessed by incorporating prediction
accuracy evaluation metric, , which is
defined in the following mathematical equation, (1):
(1)
Where,
, are the instances, which are
classified correct as positives, and,
, are the
instances, which are classified correct as negatives.
In addition, , are the instances, which are
classified false are positives, and, , are the
instances, that are classified false as negatives. A
low value of means a weak classifier while a high
value of a indicates an efficient statistical learning
classifier. Concretely, experimental assessment
based on the defined statistical quantities of: (1)
, (2)
, (3) , and (4) , which
compose the prediction accuracy evaluation metric’s
experimental value, achieve to express the data
sources’ dynamics and explain the observed optimal
results.
4.2.2 Correctly Classified Instances
In statistical machine learning, it is common to
express prediction accuracy as a percentage thus
observed results being more easily interpreted and
presented. Concretely, it is used the term correctly
classified instances, , which is
defined according to the following mathematical
equation, (2):
(2)
Where, a value close to means that the
classification algorithm is not efficient, while a
value close to indicates that the statistical
algorithm is able to classify instances optimally.
4.2.3 Confusion Matrix
We also evaluated the adopted statistical
classification algorithm with the confusion matrix
evaluation metric. Confusion matrix is a special
form of matrix, which in the case of a multiclass
classification of 5 classes, (i.e., Class 1: Aris, Class
2: Thouria, Class 3: Verga, Class 4: Arfara, and
Class 5: Meligalas) has the following encoded form,
as described in Table 1.
Where, “A” quantity depicts the number of Class
1 instances, which are classified correctly as
instances of Class 1. “B” quantity depicts the
number of Class 1 instances, which are falsely
classified as instances of Class 2. “C” quantity
depicts the number of Class 1 instances, which are
falsely classified as instances of Class 3. “D”
quantity depicts the number of Class 1 instances,
which are falsely classified as instances of Class 4.
“E” quantity depicts the number of Class 1
instances, which are falsely classified as instances of
Class 5. The same holds for the rest elements of the
confusion matrix. A given classification model is
considered efficient if it maximizes the elements of
the main diagonal of the confusion matrix (i.e., “A”,
“G”, “M”, “S”, and “Y”) and minimizes the other
elements. A confusion matrix is incorporated in
machine learning evaluation methodology to
support efficiently and explain in deep detail the
statistical nature of output experimental results
observed by the prediction accuracy evaluation
metric.
5 Experiments and Results
The data model, which is based on fused data values
is used to perform certain experiments and observe
derived results. An experimental setup is necessary
to formulate the experimental phase with certain
evaluation methods and metrics and observe the
results of the current research effort.
5.1 Experimental Setup
Specific parameters are incorporated to set up the
experimental process. Concretely, it is defined as
the number of classes, which is assigned to each
data sample instance. Intuitively, predictive
attributes used to describe a certain class are defined
accordingly. Subsequently, a certain statistical
machine learning algorithm should be adopted to
perform the experiments and observe the results.
5.1.1 Multiclass Classification
Since the number of classes is 5 this classification
process is characterized as a multiclass classification
Class 1
Class 2
Class 3
Class 4
Class 5
Classified as
A
B
C
D
E
Class 1
F
G
H
I
J
Class 2
K
L
M
N
O
Class 3
P
Q
R
S
T
Class 4
U
V
W
X
Y
Class 5
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2024.20.15
Theodoros Anagnostopoulos, Ioakeim Spiliopoulos
E-ISSN: 2224-3496
142
Volume 20, 2024
problem. Specifically, 5 classes are defined as
follows: (1) Class 1: Aris, (2) Class 2: Thouria, (3)
Class 3: Verga, (4) Class 4: Arfara, and (5) Class 5:
Meligalas. Concretely the number of predictive
attributes is also 5, which are characterized as
follows: (1) Predictive Attribute 1: ‘tocopherols’,
(2) Predictive Attribute 2: ‘phenolic compounds’,
(3) Predictive Attribute 3: ‘oxidation products of
triglycerides’, (4) Predictive Attribute 4: ‘oxidation
products of tocopherols’, and (5) Predictive
Attribute 5: ‘chlorophylls’. The number of data
sample instances is 29, which have the following
distribution per class: (1) 2 samples from Class 1
i.e., Aris, (2) 2 samples from Class 2, i.e., Thouria,
(3) 7 samples from Class 3, i.e., Verga, (4) 15
samples from Class 4, i.e., Arfara, and (5) 3 samples
from Class 5, i.e., Meligalas.
5.1.2 Logistic Statistical Learning Algorithm
To select the optimum statistical learning algorithm
that is effective in this multiclass classification
problem we experimented with several statistical
learning classifiers available in the Weka machine
learning software, [40]. Intuitively, the machine
learning algorithm, which has optimal predictive
behavior emerged to be a Logistic statistical
learning algorithm (i.e., the implementation of the
Logistic Regression algorithm in Weka machine
learning software) thus it is adopted for further
experimentation to observe the derived results of the
current research study.
5.2 Derived Results
To evaluate the experimental phase there is a need
to define a specific evaluation method (i.e., 10-fold
cross-validation) and metrics used to assess the
efficiency of the adopted statistical learning
algorithm, which in this case is the logistic
statistical machine learning algorithm. Concretely,
based on certain evaluation parameters specific
derived results are observed, which define the
effectiveness of the incorporated experimental setup
adopted in the current research effort. Intuitively, to
understand the observed results and be able to
explain the research effort’s findings it is significant
to use the incorporated evaluation method and
evaluation metrics. Such knowledge would reveal
the inherent complexity that exists in the provided
data sources aiming to observe optimal results for
the adopted machine learning algorithm.
5.2.1 Observed Prediction Accuracy
The evaluation method incorporated to evaluate the
adopted machine learning multiclass classification
algorithm is 10-fold cross-validation. According to
this evaluation method observed prediction accuracy
is: , which is a high value for prediction
accuracy thus proving that the adopted statistical
learning algorithm is suitable for the examined
multiclass classification problem. Concretely, the
high value observed for the prediction accuracy
enables the adopted machine learning algorithm to
be incorporated for similar use in new unseen olive
oil instances in a further future research that might
extend the potentiality of the current research effort
to the geographical region of interest.
5.2.2 Observed Correctly Classified Instances
According to the evaluation method of 10-fold
cross-validation correctly classified instances it
occurred to be: , which indicated that
the selected statistical machine learning algorithm is
an optimal choice for the examined classification
problem.
5.2.3 Observed Confusion Matrix
Confusion matrix results as derived based on a 10-
fold validation evaluation method for the examined
multiclass classification problem. Derived results
are presented in Table 2.
Table 2. Confusion matrix observed results
It can be observed that most of the classified
instances are located in the main diagonal of Table
2. Specifically, the quantity of elements in the main
diagonal depicts the significant number of certain
instances, which are correctly classified. Concretely,
such an optimal prediction behavior indicates a
robust classification algorithm for the examined
multiclass classification problem. Such a detailed
confusion matrix enables the observation of
experimental results in deep detail thus being able to
assess the efficiency of the adopted machine
learning algorithm for predicting PDO Kalamata
olive oil in other provided experimental instances.
6 Discussion
Problem definition indicates a multiclass
classification problem of 5 discrete classes, with 5
separate predictive attributes and a total of 29
sample instances based on the local geographic
origins of the Koroneiki olive oil variety, which is
Class 1
Class 2
Class 3
Class 4
Class 5
Classified as
2
0
0
0
0
Class 1
0
2
0
0
0
Class 2
1
0
6
0
0
Class 3
0
0
0
15
0
Class 4
0
0
0
0
3
Class 5
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2024.20.15
Theodoros Anagnostopoulos, Ioakeim Spiliopoulos
E-ISSN: 2224-3496
143
Volume 20, 2024
cultivated in the smart city of Kalamata in the
Messenia region, Greece. Subsequently, the current
research effort has achieved high values of the
observed results based on certain evaluation metrics,
which indicate the robustness of the examined
evaluation parameters. Intuitively, the current
research study has significant strengths as well as
certain weaknesses, which should be presented with
regard to a complete methodological research frame.
6.1 Weaknesses of the Study
Initial data as measured by the synchronous
photoluminescence spectra IoT device are
characterized as primitive raw data values, which
should be further processed to enter a statistical
machine learning algorithm to be evaluated
properly. Concretely, visualizing initial data sources
results in a complex plot, where there is vagueness
in distinguishing the adopted 5 classes of the initial
data source. Intuitively, such inefficiency results in
limited evaluation capability based on the available
initial data. Subsequently, classes get tangled up
with each other thus making an inference
assumption difficult to be applied. Exploitation of
visualized initial data is not suitable for further
experimentation in the current form. A fusion
process is required in the initial data to remove
outliers and missing values before further
processing.
6.2 Strengths of the Study
The data fusion process adopted in the current
research study eliminates the vagueness of the
adopted 5 data classes. Concretely, fused data
enabled the emergence of 5 discrete predictive
attributes, which aim to face the vagueness of the
initial data. Specifically, by visualizing fused data it
is proved that the classes and the predictive
attributes are distinguished easily, thus being able to
proceed with further experimentation. Intuitively,
the adopted evaluation method and metrics have
proved to be effective in defining optimal derived
results. Subsequently, the selection of a Logistic
statistical machine learning classifier emerged to be
an efficient solution to the multiclass classification
problem. Concretely, the adopted classification
algorithm was able to predict different classes based
on the fused data sources. Such effectiveness
enabled the capability of distinguishing the origin of
PDO Kalamata olive oil produced in specific local
areas in the rural areas of the Kalamata smart city.
7 Conclusions and Future Work
PDO Kalamata olive oil is an extra virgin olive oil
produced in the province of Messenia in
southeastern Greece (the name stands for capital
city Kalamata). Because of different soil
composition, microclimates, and agronomic factors
olive oil from different areas of Messenia has
diverse characteristics, although within the limits
described by council regulation (EC) No 510/2006.
Adopted synchronous photoluminescence spectra of
olive oils IoT device can specify the different
origins of PDO Kalamata olive oil. In this research
effort, we use statistical machine learning
algorithms to classify several geographic origins.
Evaluation of the statistical models are based on
certain methods and metrics, which have proved to
be promising for differentiating origins thus
enabling olive oil companies to choose PDO
Kalamata olive oil from a specific area of Messenia
with superior characteristics.
According to our research outcomes, future work
should mainly focus on the incorporation of more
detailed input measurement data sources based on
improvements in synchronous photoluminescence
spectra IoT-enabled technology, thus providing a
more robust input to the selected statistical
classification algorithm. Concretely, data fusion
techniques should be reapplied on the more detailed
initial data sources to input several statistical
learning algorithms, which might result in more
effective results. Intuitively, current research could
be further used in more detail to verify
authentication and to detect adulteration of olive oils
with protected designation of origin (PDO) thus
facing the fraud problem occurring in the olive oil
trade. Intuitively, the current research effort focuses
on a specific olive oil variety within a limited
geographic region, while future research should also
focus on the validation of the proposed
methodology to other olive oil varieties and
production areas within the Messenia region and/or
in other geographic regions of Greece that are
popular for the quality of their olive oil production.
References:
[1] S. Paiho, P. Tuominen, J. Rokman, M.
Ylikerala, J. Pajula, and H. Siikavirta,
“Opportunities of collected city data for smart
cities”, IET Smart Cities, Volume 4, Issue 4,
2022, pages 275 – 291.
[2] J. L. D. Boer, and B. Erickson, “Setting the
Record Straight on Precision, Agriculture
Adoption”, Agronomy Journal, Volume 111,
Issue 4, 2019, pages 1535 – 2139.
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2024.20.15
Theodoros Anagnostopoulos, Ioakeim Spiliopoulos
E-ISSN: 2224-3496
144
Volume 20, 2024
[3] X. Miao, J. Ma, X. Miu, H. Zhang, Y. Geng,
W. Hu, Y. Deng, and N. Li, “Integrated
transcriptome and proteome analysis the
molecular mechanisms of nutritional quality
in ‘Chenggu-32’ and ‘Koroneiki’ olives fruits
(Olea europaea L.)”, Journal of Plant
Physiology, Volume 288, Issue 154072, 2023,
pages 1 – 12.
[4] L. Trabelsi, B. Ncube, A. B. Hassena, M.
Zouairi, F. B. Amar, and K. Gargouri,
“Comparative study of productive
performance of two olive oil cultivars
Chemlali Sfax and Koroneiki under arid
conditions”, South African Journal of Botany,
Volume 154, Issue 1, 2023, pages 356 – 364.
[5] A. Issa, M. E. Riachy, C. B. Mitri, J. Doumit,
W. Skaff, and L. Karam, “Influence of
geographical origin, harvesting time and
processing system on the characteristics of
olive-mill wastewater: A step toward reducing
the environmental impact of the olive oil
sector”, Environmental Technology &
Innovation, Volume 32, Issue 103365, 2023,
pages 1 – 12.
[6] R. Aparicio, M. T. Morales, R. A. Ruiz, N.
Tena, and D. L. G. González, “Authenticity of
olive oil: Mapping and comparing official
methods and promising alternatives”, Food
Research International, Volume 54, Issue 2,
2013, pages 2025 – 2038.
[7] D. I. Ellis, H. Muhamadali, S. A. Haughey, C.
T. Elliott, and R. Goodacre, “Point–and–
shoot: Rapid quantitative detection methods
for on–site food fraud analysis–moving out of
the laboratory and into the food supply chain”,
Analytical Methods, Volume 7, Issue 22,
2015, pages 9375 – 9716.
[8] P. Rajak, A. Ganguly, S, Adhikary, and S.
Bhattacharya, “Internet of Things and smart
sensors in agriculture: Scopes and
challenges”, Journal of Agriculture and Food
Research, Volume 14, Issue 100776, 2023,
pages 1 – 13.
[9] J. Krause, H. Gruger, L. Gebauer, X. Zheng,
J. Knobbe, T. Pgner, A. Kicherer, R. Gruna,
T. Langle, and J. Beyerer, “Smart
Spectrometer–Embedded Optical
Spectroscopy for Applications in Agriculture
and Industry”, Sensors, Volume 21, Issue 13,
2021, pages 1 – 18.
[10] R. Karoui, and C. Blecker, “Fluorescence
Spectroscopy Measurement for Quality
Assessment of Food Systems – a Review”,
Food Bioprocess Technology, Volume 4,
Issue 1, 2011, pages 364 – 386.
[11] S. Khani, J. B. Ghasemi, and Z. P. Vanak,
“Development of computer vision system for
classification of olive oil samples with
different harvesting years and estimation of
chlorophyll and carotenoid contents: A
comparison of the proposed method’s
efficiency with UV-Vis spectroscopy”,
Journal of Food Composition and Analysis,
Volume 129, Issue 106078, 2024, pages 1 –
42.
[12] S. K. Drakopoulou, A. S. Kritikou, C.
Baessmann, and N. Thomaidis, “Untargeted
4D-metabolomics using Trapped Ion Mobility
combined with LC-HRMS in extra virgin
olive oil adulteration study with lower-quality
olive oils”, Food Chemistry, Volume 434,
Issue 137410, 2024, pages 1 – 9.
[13] M. E. Schiano, F. Sodano, C. Cassiano, E.
Magli, S. Seccia, M. G, Rimoli, and S.
Albrizio, “Monitoring of seven pesticide
residues by LC-MS/MS in extra virgin olive
oil samples and risk assessment for
consumers”, Food Chemistry, Volume 442,
Issue 138498, 2024, pages 1 – 8.
[14] R. Reda, T. Saffaj, I. Bouzida, O. Saidi, M.
Belgrir, B. Lakssir, and E. M. E. Hadrami,
“Optimized variable selection and machine
learning models for olive oil quality
assessment using portable near infrared
spectroscopy”, Spectrochimica Acta Part A:
Molecular and Biomolecular Spectroscopy,
Volume 303, Issue 123213, 2023, pages. 1 –
11.
[15] K. D. T. M. Milanez, T. C.A. Nóbrega, D. S.
Nascimento, M. Insausti, B. S. F. Band, and
M. J. C. Pontes, “Multivariate modeling for
detecting adulteration of extra virgin olive oil
with soybean oil using fluorescence and UV–
Vis spectroscopies: A preliminary approach”,
LWT – Food Science and Technology,
Volume 85, Issue 1, 2017, pages 9 – 15.
[16] R. A. Santos, J. C. Cancilla, A. P. Pérez, A.
Moral, and J. S. Torrecilla, “Quantifying
binary and ternary mixtures of monovarietal
extra virgin olive oils with UV–vis absorption
and chemometrics”, Sensors and Actuators B:
Chemical, Volume 234, Issue 1, 2016, pages
115 – 121.
[17] I. D. Merás, J. D. Manzano, D. A. Rodríguez,
and A. M. Peña, “Detection and quantification
of extra virgin olive oil adulteration by means
of autofluorescence excitation–emission
profiles combined with multi–way
classification”, Talanta, Volume 178, Issue 1,
2018, pages 751 – 762.
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2024.20.15
Theodoros Anagnostopoulos, Ioakeim Spiliopoulos
E-ISSN: 2224-3496
145
Volume 20, 2024
[18] Y. Li, T. Fang, S. Zhu, F. Huang, Z. Chen,
and Y. Wang, “Detection of olive oil
adulteration with waste cooking oil via Raman
spectroscopy combined with iPLS and
SiPLS”, Spectrochimica Acta Part A:
Molecular Biomolecular Spectroscopy,
Volume 189, Issue 1, 2018, pages 37 – 43.
[19] F. D. Girolamo, A. Masotti, I. Lante, M.
Scapaticci, C. D. Calvano, C. Zambonin, M.
Muraca, and L. A. Putignani, “Simple and
effective mass spectrometric approach to
identify the adulteration of the mediterranean
diet component extra–virgin olive oil with
corn oil”, International Journal of Molecular
Sciences, Volume 16, Issue 9, 2015, pages
20896 – 20912.
[20] A. Rotondo, L. Mannina, and A. Salvo,
“Multiple Assignment Recovered Analysis
(MARA) NMR for a Direct Food Labeling:
The Case Study of Olive Oils”, Food
Analytical Methods, Volume 12, Issue 1,
2019, pages 1238 – 1245, DOI:
10.1007/s12161-019-01460-4.
[21] M. M. Mossoba, H. Azizian, A. R. F. Kia, S.
R. Karunathilaka, and J. K. G. Kramer, “First
Application of Newly Developed FT–NIR
Spectroscopic Methodology to Predict
Authenticity of Extra Virgin Olive Oil Retail
Products in the USA”, Lipids, Volume 52,
Issue 5, 2017, pages 443 – 455, DOI:
10.1007/s11745-017-4250-5
[22] H. Zaroual, C. Chene, E. M. E. Hadrami, and
R. Karoui, “Comparison of four classification
statistical methods for characterizing virgin
olive oil quality storage up to 18 months”,
Food Chemistry, Volume 370, Issue 131009,
2022, pages 1 – 16.
[23] X. Wu, S. Gao, Y. Niu, Z. Zhao, B. Xu, R.
Ma, H. Liu, and Y. Zhang, “Identification of
olive oil in vegetable blend oil by one-
dimensional convolutional neural network
combined with Raman spectroscopy”, Journal
of Food Composition and Analysis, Volume
108, Issue 104396, 2022, pages 1 – 7.
[24] I. H. A. S. Barros, L. S. Paixao, M. H. C.
Nascimento, V. J. Lacerda, P. R. Figueiras,
and W. Romao, “Use of portable Raman
spectroscopy in the quality control of exrtra
virgin olive oil and adulterated compound
oils”, Vibrational Spectroscopy, Volume 116,
Issue 103299, 2021, pages 1 – 10.
[25] L.Cecchi, M. Migliorini, E. Giambanelli, A.
Rosseti, A, Cane, N. Mulinacci, and F.
Melani, “Authentication of the geographical
origin of virgin olive oils from the main
worldwide producing countries: A new
combination of HS-SPME-GC-MS analysis of
volatile compounds and chemometrics applied
to 1217 samples”, Food Control, Volume 112,
Issue 107156, 2020, pages 1 – 10.
[26] D. Stefas, N. Gyftokostas, P. Kourelias, E.
Nanou, V. Kokkinos, C. Bouras and S.
Couris, “Discrimination of olive oils based on
the olive cultivar origin by machine learning
employing the fusion of emission and
absorption spectroscopic data”, Food Control,
Volume 130, Issue 108318, 2021, pages 1 – 8.
[27] V. Rotich, D. F. A. Riza, F. Giametta, T.
Suzuki, Y. Ogawa, and N. Kondo, “Thermal
oxidation assessment of Italian extra virgin
olive oil using an UltraViolet (UV) induced
fluorescence imaging system”,
Spectrochimica Acta Part A: Molecular and
Biomolecular Spectroscopy, Volume 237,
Issue 118373, 2020, pages 1 – 8.
[28] A. Bagnall, L. Davis, J. Hills, and J. Lines,
“Transformation Based Ensembles for Time
Series Classification”, Proceedings of the
2012 SIAM International Conference on Data
Mining (SDM), Anaheim, California, USA,
April 26 – 28, 2012, pages 307 – 318.
[29] H. S. Tapp, M. Defernez, and E. K. Kemsley,
“FTIR Spectroscopy and Multivariate
Analysis Can Distinguish the Geographic
Origins of Extra Virgin Olive Oils”, Journal
of Agricultural and Food Chemistry, Volume
51, Issue 21, 2003, pages 6110 – 6115, DOI:
10.1021/jf030232s.
[30] P. D. L. Mata, A. D. Vidal, J. M. B. Sendra,
A. R. Medina, L. C. Rodriguez, and M. J. A.
Canada, “Olive oil assessment in edible oil
blends by means of ATR-FTIR and
chemometrics”, Food Control, Volume 23,
Issue 2, 2012, pages 449 – 455.
[31] E. Sikorska, I. Khmelinskii, and M. Sikorski,
“Analysis of Olive Oils by Fluorescence
Spectroscopy: Methods and Applications”,
InTech, Volume 1, Issue 1, 2012, pages 63 –
88, DOI: 10.5772/30676.
[32] E. Sikorska, T. Gorecki, I. V. Khmelinskii, M.
Sikorski, and J. Koziol, “Classification of
edible oils using synchronous scanning
fluorescence spectroscopy”, Food Chemistry,
Volume 89, Issue 2, 2005, pages 217 – 225.
[33] E. Sikorska, A. G. Swiglo, I. Khmelinskii, and
M. Sirorski, “Synchronous Fluorescence of
Edible Vegetable Oils. Quantification of
Tocopherols”, Journal of Agriculture and
Food Chemistry, Volume 53, issue 18, 2005,
pages 6988 – 6994, DOI: 10.1021/jf0507285.
WSEAS TRANSACTIONS on ENVIRONMENT and DEVELOPMENT
DOI: 10.37394/232015.2024.20.15
Theodoros Anagnostopoulos, Ioakeim Spiliopoulos
E-ISSN: 2224-3496
146
Volume 20, 2024
[34] N. Dupuy, Y. L. Dreau, D. Ollivier, J. Artaud,
C. Pinatel, and J. Kister, “Origin of French
Virgin Olive Oil Registered Designation of
Origins Predicted by Chemometric Analysis
of Synchronous Excitation-Emission
Fluorescence Spectra”, Journal of
Agricultural and Food Chemistry, Volume 53,
Issue 24, 2005, pages 9361 – 9368.
[35] R. M. Maggio, E. Valli, A. Bendini, A. M. G.
Caravaca, T. G. Toschi, and L. Cerretani, “A
spectroscopic and chemometric study of
virgin olive oils subjected to thermal stress”,
Food Chemistry, Volume 127, Issue 1, 2011,
pages 216 – 221.
[36] E. Bertran, M. Blance, J. Coello, H. Iturriaga,
S. Maspoch, and I. Montoliu, “Near infrared
spectrometry and pattern recognition as
screening methods for the authentication of
virgin olive oils of very close geographical
origins”, Journal of Near Infrared
Spectroscopy, Volume 8, Issue 1, 2000, pages
45 – 52.
[37] A. L. Prieto, N. Tena, R. A. Ruiz, D. L. G.
Gonzalez, and E. Sikorska, “Monitoring
Virgin Olive Oil Shelf-Life by Fluorescence
Spectroscopy and Sensory Characteristics: A
Multidimensional Study Carried Out under
Simulated Market Conditions”, Foods,
Volume 9, Issue 12, 2020, pages 1 – 20, DOI:
10.3390/foods9121846
[38] The LS-55 and LS-45 Fluorescence
Spectrofluorometers, Perkin Elmer, [Online].
https://resources.perkinelmer.com/lab-
solutions/resources/docs/BRO_LS-55andLS-
45FluorescenceSpectrophotometer.pdf
(Accessed Date: February 26, 2024).
[39] E. Frank, M. A. Hall, and I. H. Witten, The
WEKA Workbench. Online Appendix for
“Data Mining: Practical Machine Learning
Tools and Techniques”, Morgan Kaufmann,
Fourth Edition, 2016.
[40] M. Hall, E. Frank, G. Holmes, B. Pfahringer,
P. Reutemann, and I. H. Witten, “The WEKA
Data Mining Software: An Update”, SIGKDD
Explorations, Volume 11, Issue 1, 2009,
pages 10 – 18.
Contribution of Individual Authors to the
Creation of a Scientific Article (Ghostwriting
Policy)
The authors equally contributed to the present
research, at all stages from the formulation of the
problem to the final findings and solution.
Sources of Funding for Research Presented in a
Scientific Article or Scientific Article Itself
No funding was received for conducting this study.
Conflict of Interest
The authors have no conflicts 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: 10.37394/232015.2024.20.15
Theodoros Anagnostopoulos, Ioakeim Spiliopoulos
E-ISSN: 2224-3496
147
Volume 20, 2024
